Mohammad Imran Khan Mewati teaches grades 6-12 at a school in rural India. His students are largely from economically disadvantaged families, and the school itself has limited resources. But the biggest problem, he says, is absenteeism. 

“If a student is not coming into your class, how are you going to teach?” says Mewati, a teacher for 26 years. “That’s why I’m using technology in my classroom and outside the classroom, so that they can learn a little bit using their smartphones.” 

MIT Open Learning’s free educational resources have been a boon for Mewati as he develops Hindi-language digital resources for his students. 

“I am a self-taught app developer,” he explains. “MIT Open Learning has had a deep and practical impact on my professional life as a teacher. Many concepts I learned influenced how I design digital learning activities, simple educational games, and classroom strategies. My students may not know they are indirectly benefiting from MIT, but they are.”

Through MIT Open Learning, Mewati has used in his classroom OpenCourseWare’s free, online library of educational resources from more than 2,500 courses spanning the MIT undergraduate and graduate curriculum. Learners can browse content at their own pace, watch lectures, read course notes, and hear from faculty experts. All materials can be downloaded for offline use, and the website is fully responsive for smartphone use. These materials are also available on MIT Learn, an AI-enabled platform for all of MIT’s lifelong learning opportunities.

Mewati started using OpenCourseWare resources in the early 2010s and cites programming courses as the most useful. He dove deep into Introduction to CS and Programming Using Python, Introduction to C and C++, and Introduction to Programming Using Java. Introduction to Computational Thinking helped him bring together problem-solving approaches from mathematics and computer science as he built apps.  

Closing the gap with technology

Mewati’s school is located 15 miles from the city of Alwar in Rajasthan, a state in northwestern India. Most of the students do not have access to desktop or laptop computers, but the majority live in a home where at least one person has a smartphone. Taking advantage of this technology, Mewati creates classroom groups on WhatsApp so that he can share resources with his students, regardless of whether they can make it to class. 

Mewati began incorporating technology into his teaching in the early 2010s, when he had to engage 180 students in a lesson about the moon landing and Neil Armstrong. Mewati created a simple HTML page that included many iconic images — the American flag planted on the moon, the Apollo 11 spacecraft, the footprint on the moon’s surface — with Hindi explanations. Once he created it, he could use it again and again. When he tested students on what they’d learned, they got better results than when he’d taught the material using his previous approach. 

“This is the incident that confirmed to me that technology can play an important role in the lives of the students, and particularly for the rural students, for the students who do not have equal opportunities,” says Mewati. “Technology is the equalizer.”

That belief has driven Mewati’s efforts to build his school’s technological resources. Through crowdfunding, he secured 15 used computers to create a computer lab, and teachers now share the responsibility for creating mobile hotspots so students can connect to the internet. Mewati’s mobile apps provide additional opportunities for students to explore topics in greater depth or catch up on lessons they may have missed.

Referring to his apps as his favorite topic, Mewati explains that he has developed several types to meet a variety of goals. Some allow students to play games that develop their math skills, while others include syllabi, reading recommendations, and class notes. Other apps help students prepare for exams required for government jobs. He also builds apps for audiences beyond his school, such as an app that provides Hindi-language resources related to maternal health, a topic he says is not openly discussed in India. The app has been popular, he explains, because it provides honest information that people can view privately.

“If I see an issue, I think, ‘Yes, let’s create an app,’” he says. 

Collaboration, open sharing, and lifelong learning

Mewati shares the apps he has created, and the MIT Open Learning resources that support him, with a network of teachers across India. He has connected with other educators through India’s National Teacher Awards and earlier this year, he traveled to Dubai for the Global Teacher Prize.

“We share things with each other,” he says. “We have a huge group — more than 1,000 teachers connected across India. So, if we see resources, or useful things for a class, we’ll share. All the time, I talk about open-source materials, like MIT courses, for educational purposes.”

Through this culture of collaboration and open sharing, Mewati is able to bring new learning opportunities to his students. The apps he has built — made possible by what he has learned through MIT Open Learning — help extend access to educational resources beyond the classroom. As an educator, he says he’s happy about that. But he is also a learner, and it’s his journey as a learner that he wants people to know. 

“I am from a rural area. My parents are not educated at all. And I am a Fulbright Scholar. I can reach the Global Teacher Prize stage. The reason is simple,” he says. “It is because I continued my learning, whenever possible, with the use of technology and availability of free courses from MIT.”

To anyone who is curious, who wants to learn, to push themselves or build something new, he says that with MIT Open Learning, the resources are out there.

“We should use it, we should grab it, and we should share it as much as we can,” says Mewati. “Because that’s how humanity can flourish.”

MIT’s Kavli Institute for Astrophysics and Space Research (MKI) is launching a pilot artist-in-residence program to facilitate cross-disciplinary dialogue between art, science, and the public. 

MKI is a world-leading institution for research in astrophysics, combining more than 60 years of expertise in space and ground-based instrumentation development with the intellectual energy of MIT’s faculty, research and technical staff, and students in the departments of Aeronautical and Astronautical Engineering; Earth, Atmospheric and Planetary Sciences; and Physics.

During the 2026-27 academic year, internationally acclaimed ultra-contemporary artist Amy Karle will work as the program’s inaugural artist-in-residence alongside MKI researchers to explore the research and processes behind cutting-edge astrophysical discoveries and instrumentation, and to translate this experience into an immersive, multimedia installation available for public display beginning in early 2028. Karle is known for her work as an artist, designer, and researcher whose projects explore how science and technology shape humanity, evolution, and the future across scales and systems, from cells to cosmos. 

“We are excited to work with Amy in this collaborative environment” says MKI Director Robert Simcoe, the Bruno B. Rossi Professor of Experimental Physics at MIT. “Her approach is unlike anything we have previously experienced at MKI and presents many opportunities to challenge the way we, as scientists and engineers, think about our study of the universe. At the same time, the resulting artwork will be shaped by the deep research we do, and the wide-ranging scientific and technical perspectives of the MKI community.” 

Karle’s proposal, which envisions astrophysical research and data as a co-creative experience toward embodied understanding of cosmic phenomena and touches on themes of scientific observation, signals, and inference, was selected by an interdisciplinary committee of astronomers, museum curators, and art-science practitioners. Reviewers praised Karle’s ambitious-yet-grounded approach to engagement, her attention to audience experience, her unique approach to science communication through art and technology, and the collaborative potential of her artistic vision. 

“I am thrilled to be partnering with MKI,” says Karle, whose practice over the years has included dedicated art-science collaborations with Copernicus Science Centre, the Interstellar Foundation, and Studio Quantum, as well as multiple installations for museums, festivals, and public spaces across the globe. “My first job was at a public observatory. I still remember showing strangers Saturn’s rings through a telescope and watching awe and understanding arrive as felt experience. That has shaped my work since. What MKI does at the frontier of astrophysics, translating faint signals into knowledge through instruments, computation, and human judgment, is a profound expression of that same process. I am excited to be in dialogue with that work and with MKI scientists to create art that makes this tangible and deeply felt, inviting people into the threshold where our ways of knowing the universe reshape how we understand ourselves.”

The residency begins with a one-month exploratory period in the fall semester, centered on meetings with MKI researchers, attendance at seminars and classes, and a public presentation to the MKI community. The project will then move from conceptualization to development, shaped through continued exchange with MKI researchers and complementary independent work in Karle’s California studio throughout 2027. 

In March, Karle and selected scientific collaborators will be in residence at the Studios at MASS MoCA, a national and international residency program embedded within one of the world’s largest and liveliest museums dedicated to contemporary art. During their time in residence, Karle and collaborators will test ideas, exchange knowledge, and engage with a multidisciplinary cohort of 16 other artists from across the globe.

“MASS MoCA [the Massachusetts Museum of Contemporary Art] is pleased to be part of MKI’s artist-in-residence program and to contribute to the meaningful exchange between art and science,” says Susan Cross, MASS MoCA director of curatorial affairs. “We look forward to welcoming artist Amy Karle and collaborators from MIT’s Kavli Institute for Astrophysics and Space Research to our campus, and to the Studios at MASS MoCA.”

The residency is supported by the Kavli Foundation’s Kavli Innovation Fund. The initiative seeks to develop new modes of public engagement with astrophysical research and discovery, and deepen emotional connections across the interplay of science and art.

“We are grateful for the Kavli Foundation’s support,” says Simcoe, “as it allows us to push boundaries and engage new audiences in the wonder of the universe and the process of science.” 

Karle’s work has been exhibited internationally at institutions including Centre Pompidou, Mori Art Museum, the Smithsonian Institution, the Museum of Modern Art, Ars Electronica, ArtScience Museum, Triennale Milano, and the Victoria and Albert Museum, with works on the moon and in space. She collaborates with and presents at scientific, technological, and cultural institutions including NASA, CERN, SLAC National Accelerator Laboratory, Autodesk, HP Labs, and NVIDIA. 

She was honored as one of BBC’s 100 Most Inspiring and Influential Women, a Pioneer in Design, and one of the Most Influential Women in 3D Printing. Karle also served as an American Arts Incubator U.S. Department of State artist diplomat. Her first job was at a public observatory, where she began asking fundamental questions about space and witnessing the wonder it can awaken in people, an early experience that continues to inspire her to create works that allow people to feel how we come to know the universe and our place within it.

To learn more about Karle's work, visit amykarle.com. As the project develops, MKI will be seeking museum and festival partners to host the installation in 2028 and beyond. 

MIT researchers have created a new computing platform that could be used to develop intelligent and adaptive next-generation electronics that can simultaneously perform multiple functions, like computing and memory, all within one extremely compact, energy-efficient device.

Such a platform opens opportunities for low-power edge computing applications, interactive medical and environmental monitoring systems, and smart robots.

The researchers accomplished this by leveraging the unique mechanical response of soft polymers at the nanoscale. A mechanical response is how a structure changes when a force is applied to it. 

They harnessed this response to create tiny mechanical devices that use reconfigurable motion to remember and process information in a way that mimics how neurons behave in the brain.

Because key computing functions are built into the intrinsic properties of the soft polymer material, the number of components needed to perform the functions are minimized, enabling a compact and versatile platform for information processing. 

“Complex and coupled nanoscale phenomena can provide tremendous opportunities for new approaches to information processing and integrating multiple functionalities, such as computing, sensing, and actuation. This could enable levels of energy efficiency, autonomy, and reconfigurability in nanoscale devices and systems that are challenging to achieve with conventional computing platforms,” says Farnaz Niroui, an associate professor of electrical engineering and computer science (EECS), a member of the Research Laboratory of Electronics (RLE), and senior author of a paper on this device. “Here, we harness the intrinsic mechanical properties of materials to engineer device-level dynamics, such that the material building blocks play a much more active role in defining device functionality than conventionally considered.”

She is joined on the paper by co-lead authors Peter Satterthwaite and Sarah Spector, EECS graduate students; as well as Jeremiah Johnson, the A. Thomas Guertin Professor of Chemistry at MIT; Maxwell Conte, a graduate student in the Department of Materials Science and Engineering; Teddy Hsieh, an EECS graduate student; postdoc Eduard Bobylev; and Srinidhi Venkatesh ’25. The research appears today in Science Advances. 

Bioinspired computation

Biological systems can leverage physical changes, like motion or deformation, to process information efficiently and without needing access to a central controller. 

For instance, an octopus has a highly distributed nervous systems, with about two-thirds of its neurons spread throughout its arms. This allows the octopus to sense and process information about its environment locally and generate responses without requiring access to the central brain.

As an example, an octopus can mechanically change the color cells in its skin, enabling it to go through a rapid and context-specific camouflage process.

“You can think of an octopus as continuous computing matter, with computing, memory, sensing, and actuation distributed throughout its body,” Niroui adds.

Inspired by such performance, the researchers sought to develop a platform that can compute using mechanical transformations at the nanoscale. In mechanical computing, calculations are performed through physical transformations like movement and compression. 

While bioinspired mechanical computing platforms have been developed at the micro and macro scales, the MIT researchers shrunk their device to the nanoscale. At this scale, even minute mechanical transformations can lead to drastic changes in a material’s properties. This can enable complex computing in an energy-efficient platform.

But achieving the reversible nanomechanical transformations needed for such computing is a fundamental challenge. When two surfaces come very close, they experience strong adhesive forces that pull the surfaces together, making them impossible to unstick. 

To overcome this fundamental challenge, the researchers built a device with a super-thin film of the soft polymer polydimethylsiloxane (PDMS) sandwiched between two metal electrodes. This soft spacer balances the adhesive forces between the two metal surfaces, keeping the electrodes from crashing together in an irreversible way.

“The soft material in serves as a ‘nano-spring,’ to help balance the forces to achieve nanoscale mechanical reconfiguration in a controlled and reversible manner,” Niroui explains.

When the researchers apply a voltage to the device, the two metal plates attract to one another, compressing the soft material and altering the electrical current flowing through the device. 

“PDMS is viscoelastic, which means that after being compressed, it takes time to return to its original state. This allows the devices to dynamically remember the history of forces and voltages applied to them, and convert that history into an electrical response,” says Satterthwaite.

They researchers used this performance to demonstrate an artificial neuron.

Brain-inspired information processing

In the brain, each neuron accumulates an electrical charge a little bit at a time until it reaches a threshold and fires, passing information to other neurons in the network. 

The researchers’ device mirrors this behavior. As voltage is applied over time, it accumulates stimulus as the electrodes gradually compress the PDMS. After crossing a threshold, it “fires” like a neuron before relaxing back to its original state.

“We have this complex functionality, which is the basis of biological computing, all contained in one nanoscale device,” Satterthwaite says.

Since computing and memory are incorporated within a single device with no need for external components, like capacitors or complex circuitry, this platform can achieve high energy efficiency with a small footprint. 

“The performance highly relies on the memory introduced using the soft polymer. We can intentionally engineer this over a large design space to meet the requirements of the desired applications,” Spector says.

The device can also be compatible with biological systems, Spector adds. For instance, it could be useful in applications like smart prosthetics that can rapidly process tactile data or low-power wearable patches that collect and analyze health indicators in real-time.

In the future, the researchers want to expand this work to further integrate sensing with computing and memory to realize nanomechanical computing matter with applications in intelligent and adaptive systems. 

This work was funded, in part, by the U.S. Defense Advanced Research Projects Agency (DARPA), the U.S. National Science Foundation (NSF), an MIT EECS MathWorks Fellowship, and the Netherlands Organization for Scientific Research. Device fabrication was carried out, in part, using MIT.nano facilities.

Researchers created a new technique that accurately and rapidly matches X-rays captured during surgery with a patient’s preoperative 3D medical scan. This method could make it easier for clinicians to precisely pilot minimally invasive surgical tools, leading to faster and safer procedures.

Clinicians perform many minimally invasive surgeries using real-time X-rays to help them steer devices like catheters and endoscopes through tiny incisions. But since X-rays are flat images, it can be challenging to determine exactly where surgical tools are located and oriented within the patient’s body, increasing the risk of complications.

To help localize surgical devices, clinicians may manually align X-rays with preoperative 3D medical images, such as CT scans or MRIs. Artificial intelligence tools designed to streamline this process struggle to align images robustly for all patients, making them infeasible in practice.

This new system, developed by scientists and clinicians at MIT and collaborating institutions, uses an AI model that adapts to each patient in only about five minutes. The model automatically matches one patient’s X-rays with 3D scans in a matter of seconds, and with sub-millimeter precision.

Named xvr (which stands for X-ray volume registration), it outperformed existing AI methods by an order of magnitude across a wide range of patients, body parts, and medical procedures.

“A majority of Americans live more than an hour away from a center that can perform noninvasive procedures, like emergency stroke interventions. An hour in stroke time is incredibly substantial. Making these procedures easier by combining 2D and 3D information enables these types of highly specialized life-saving procedures to be more accessible to much broader parts of the population,” says Vivek Gopalakrishnan, a postdoc in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL); a recent graduate of the Harvard-MIT Program in Health Sciences and Technology; and lead author of a paper on xvr, which appears today in Nature.

He is joined on the paper by his advisor Polina Golland, the Sunlin and Priscilla Chou Professor of Electrical Engineering and Computer Science (EECS), a principal investigator in CSAIL, the leader of the Medical Vision Group, and co-senior author of the paper; and Neel Dey, a former postdoc in the Medical Vision Group who is now an investigator at Harvard Medical School and Massachusetts General Hospital as well as co-senior author on the paper. Additional co-authors include David-Dimitris Chlorogiannis, a researcher and clinician at Harvard Medical School; Andrew Abumoussa, a neurosurgeon at St. Luke’s Marion Bloch Neuroscience Institute; Anna M. Larson, a pediatric clinician at Shriners Children’s Hospital; Nazim Haouchine, an assistant professor of radiology at Harvard and Brigham and Women’s Hospital; Darren B. Orbach, a physician and scientist at Boston Children’s Hospital; and Sarah Frisken, an associate professor of radiology at Harvard.

Making X-rays more informative

In many minimally invasive surgical procedures, like angioplasty to open blocked arteries, clinicians insert instruments through a tiny incision and use a high-speed mobile X-ray scanner to generate images that allow them to visualize the procedure from any angle. 

But to guide surgical tools without accidentally damaging other tissue, clinicians must align real-time X-rays with the patient’s preoperative MRI or CT scan. This process, called registration, helps them determine where the tool is in relation to anatomical structures. 

“It takes decades of training for a clinician to become skilled enough to see grainy, 2D images and understand how everything is oriented. We want to make these 2D X-rays more informative, so it becomes safer and easier to do these life-saving procedures,” Gopalakrishnan says.

Manual registration methods are slow and burdensome, requiring the clinician to guess the position of a surgical instrument by punching numbers into a computer or clicking anatomical landmarks on a screen. 

To streamline the process, researchers are developing AI models that can predict 2D/3D registration. But people have such diverse anatomy that a model which works well for some patients may fail for others. 

A lack of high-quality annotated medical image data makes it difficult to train a deep-learning model robust enough to adapt to many patients, Gopalakrishnan says.

Rather than trying to make a machine-learning model that can be applied to all patients, the researchers built a model designed to adapt extremely well for the specific patient.

“We tailor this one specific model for this one specific patient, and it doesn’t matter if it works on other people because there will be different models for those people,” Gopalakrishnan adds.

Patient-specific machine learning

Xvr takes one patient’s preoperative 3D scan, like an MRI or CT, and uses it to generate thousands of synthetic X-rays from many angles, producing about 1,000 images each second. It uses a physics-based simulation of the X-ray process to ensure these synthetic images are realistic.

“Instead of generating data from nothing, like some types of generative AI, this physics simulation is entirely based on the CT scan or MRI from this patient. Because xvr creates patient-specific data in a purely physics-based manner, there is no room for hallucinations,” Gopalakrishnan says.

The xvr framework uses these simulated data to train an AI model that can accurately align this patient’s 2D X-rays with their 3D image scan in a matter of seconds.

But while such a registration model is highly accurate, it would take about 12 hours to train from scratch for each patient, making it impossible to deploy in an emergency. To make the process faster, the researchers used xvr to pretrain a more versatile AI system, called a foundation model, that can quickly adjust to each new patient. 

They collected whole-body 3D medical scans from more than 2,000 patients covering a wide range of ages, image modalities, and regions. Xvr used these diverse data to generate synthetic X-rays and train a foundation model to perform 2D/3D registration.

This pretrained model can adapt to a new patient in about five minutes, and performs registration with the same accuracy as if it had been trained from scratch. 

“So now you can get patient-specific accuracy but also in a very rapid time frame,” Gopalakrishnan says.

The team tested the model on the largest available dataset of real 2D/3D registrations, incorporating data from five hospitals that covered dozens of bones and organ systems in adult and pediatric patients. 

Xvr significantly outperformed other AI-based methods in accuracy and robustness, while operating fast enough for emergency surgeries. The model could also be used to improve the performance of robotic surgery technologies. 

In the future, the researchers hope to focus on making xvr faster for real-time deployment, conducting further studies to verify its reliability in additional situations, and extending the system to handle more complex scenarios, like moving body parts. 

“For the past two years, we’ve been carefully developing this algorithm and validating it. Now, we are collaborating closely with surgical robotics companies and clinical groups to turn this research into useful tools for navigation or deployment,” Gopalakrishnan says.

This work was funded, in part, but the National Institutes of Health (NIH), the MIT CSAIL-Wistron Program, the MIT-IBM Computing Research Lab, the MIT Jameel Clinic, the MIT Health and Life Sciences Collaborative, and the Chou Family Transformative Research Fund.

The annual “Demo Day” event at MIT, which marks the end of the delta v startup accelerator, fell on the 25th anniversary of the Sept. 11 attacks this year, giving MIT entrepreneurs a chance to honor the memory of those lost that day while presenting their startup progress in the program.

Each year, the event celebrates all that students achieved while working full-time on their ventures over the summer with support and guidance from the Martin Trust Center for MIT Entrepreneurship.

But the usually boisterous night started with the program’s military veterans asking for a moment of silence.

“Today is a day of remembrance, but also a day of celebration,” founder and MIT graduate student Kevin Power MAP ’25 told the audience in opening remarks. “It’s about building to create a better world. Today, we honor those lost the way we believe they would want: by being humble, taking care of each other, and building something worthy of the people who never had this chance. In this room, people are taking on the hardest problems in health care, cybersecurity, defense, robotics, and manufacturing.”

Now in its 15th year, delta v Demo Day gives MIT entrepreneurs a chance to share their work and inspire classmates to adopt the entrepreneurial mindset. The companies that presented were whittled down from an initial list of over 200, twice the amount that applied in 2025.

Across a whirlwind 90 minutes inside a jam-packed Kresge Auditorium, 13 teams presented their startups to the audience in two-minute presentations. Many shared business milestones and progress in line with what a typical company would achieve over multiple years, including customer partnerships, prototype deployments, and even revenue.

Each team received mentorship and support along with $75,000 in equity-free funding, a dramatic increase from years past. This year’s cohort featured undergraduates, graduate students, and postdocs, from across all of MIT’s schools.

“One of the things I love about delta v is it brings students from all across our community together to approach challenges with different perspectives,” Paula Hammond, dean of the MIT School of Engineering, told the audience. “Their companies are just as wide-ranging. They are working in AI, robotics, health care, aerospace, financial technology, biotech, cybersecurity, and more. At their core, they all share a desire to tackle difficult problems and improve people’s lives.”

This year the Trust Center also announced a new partner model for the delta v program, composed of over 125 leading founders from companies like HubSpot, Okta, and Kayak, along with industry experts and early-stage investors.

The event’s occurrence at the start of the semester is no coincidence: It is timed to attract the next generation of entrepreneurs on campus.

“This is my favorite day of the year,” said Bill Aulet, the managing director of the Trust Center and MIT’s Ethernet Inventors Professor of the Practice at the MIT Sloan School of Management. “Today is about building organizations that will solve the world’s most intractable problems. It’s about more than making money. These presentations will inspire you and make you proud to be a part of the MIT community.”

Artificial intelligence featured prominently in this year’s cohort of companies, which are applying the technology to solve major problems in cybersecurity and manufacturing, improve health care spending, design advanced metal parts, and more.

The company Neural Physics, for instance, is building AI models for manufacturing and other hardware applications. The company’s models are designed to accelerate product design and validation workflows for companies building things like cars, equipment, and machine parts.

“AI can build software overnight,” said co-founder and PhD candidate Mohamed Elrefaie. “AI for software has been solved. The next revolution is physical AI. Design takes too long, and it costs billions. In 1907, it took Henry Ford five years to develop the first Ford car model. Today, it still takes the Ford Motor Company five years to go from design to production. The U.S. advanced manufacturing sector loses roughly $245 billion annually due to engineer delays… [Most] of that time is spent running simulations or making engineering decisions. At Neural Physics, we are building foundation physics models to accelerate those processes.”

Another company, Cerebrus AI, has built a system for detecting when AI agents deviate from approved behavior. The solution builds a baseline of behavior for each deployed agent and monitors their activity to flag unusual behavior that could lead to problems.

“The rollout of revolutionary technology is being held up by three key questions that every executive is asking: Where are my agents? What are they doing? What do they have access to?” co-founder and MBA student Griffin Potrock said. “Security teams want to say yes, but they can’t trust what they can’t see. Cerebrus AI can help those teams.”

The company Talys uses AI agents to help health care organizations find opportunities to lower spending on things like pharmacies, operational processes, and third-party services. The company is already working with health systems and has processed $325 million in spending.

“Decades of attempts to reign in health care spending have fallen short — until now,” co-founder and MBA student Nicolas Berzin said “Why is it so hard? Analytics and dashboards give you pictures of the problem, but not the solution. Meanwhile, consultants are slow and expensive. There are thousands of spend categories, tens of thousands of procedures, and millions of items. Who knows how to save on all of these things? Imagine if you could classify every line, benchmark every item, find every substitution, triage every unprofitable case, and model every scenario across multiple contracts and thousands of procedures and categories like an expert. Talys is a margin-execution system that runs 24/7 to optimize procurement, reduce leakage, and improve case economics.”

Other delta v teams also presented impressive hardware solutions. RBT Resources presented a portable device that simplifies and speeds up blood transfusions, which could be used in hospitals and at the site of traumatic injuries like highways or battlefields.

“Transfusion at the point of injury is an extremely manual process with three key inefficiencies: They are time dependent, gravity dependent, and labor intensive,” explained CEO Anthony Capuano MBA ’26, a former U.S. Navy Seal. “Our goal at RBT Resources is to make transfusions faster and simpler for all medics.”

Gander Robotics developed a low-cost drone submarine that can be used when someone falls overboard on a ship. The hand-thrown, autonomous vessel can sense and travel to the person at sea and give them something to hold onto at the surface, all while providing rescue crews with its exact location.

So-called “man-overboard” situations are surprisingly common on military boats and cruise ships. The device was developed over two years at MIT and the Woods Hole Oceanographic Institute. “Our autonomous rescue swimmer uses a proprietary technique to search with sonar from underneath the surface, where it’s nice and calm even if there’s a storm raging above,” CEO Michael Autery MBA ’26 explained.

The other teams presenting included:

Alpaca is building an integrated ecosystem of hardware and software to allow individuals to host their own frontier AI models without a subscription.

Banzai is building an AI-powered agent to help homeowners, property managers, and asset managers diagnose home repairs faster, improve repair accuracy, and reduce maintenance costs.

Bizon Labs is building a platform for engineering lipid nanoparticles to deliver advanced medicine anywhere in the body.

Cortheon uses AI design optimization to help foundries make complex metal parts at lower cost and with the design freedom of 3D printing.

Exo AI is helping financial institutions automate back-office processes using AI-native software capable of analyzing messy data and connecting fragmented workflows.

Pixology is using agentic AI to help sales teams create visual, engaging pitch materials faster for media rights deals.

Robox is using AI to develop a design engine for physical automation inside systems integrators, robotics firms, and manufacturers.

The Trade Lab is helping importers navigate shifting tariff regulations across the globe and optimize supply chains.

Let’s agree at the outset the world is a complicated place, and social scientists have exacting jobs when it comes to measuring civic phenomena with precision. 

After all, even careful studies raise follow-up questions: How much do their findings apply in other settings? Do conclusions about politics in one country apply to other countries? If you’re studying voters in a lopsided election, will your findings apply to voters in a close election? Those questions are all a natural part of the research process.

That’s where Naoki Egami comes in. Egami is an MIT political scientist whose specialty is the methodology of research. He carefully scrutinizes, for one thing, what social scientists call “external validity,” whether the results of particular studies apply more generally.

“I always say political methodology is the field where you ask questions as a political scientist, but then you solve them like an applied statistician or an applied computer scientist,” Egami says. “You find out the underlying mathematical problems behind the empirical challenges people face, and solve them optimally.”

As it happens, Egami’s interests range widely. Years ago, before the current artificial intelligence craze, he started studying what happens when AI tools are introduced into studies. How accurate are they? How can researchers account for AI tendencies? Focusing on these and other questions has helped Egami build a broad portfolio of research, win awards, and flourish in his career. All the while, he retains interest in basic questions about politics, as well as measuring things correctly. 

“You need both perspectives,” Egami says. “If you only think about technical statistical theories, you might not work on interesting empirical problems sometimes. But if you only think about problems, you won’t really solve them optimally; you’ll solve them in an ad-hoc way. So, you really want to have both lenses.”

Egami joined MIT’s Department of Political Science as an associate professor with tenure in 2025. He is also a faculty affiliate of the Statistics and Data Science Center at the Institute for Data, Systems, and Society (IDSS).

Workshopping his career

Almost anyone who likes their job has experienced some good fortune in finding it. Egami’s case calls to mind those adages about luck being a mixture of preparation and opportunity. 

Egami grew up in Tokyo and attended the University of Tokyo. He was good at math and physics, but he also liked political philosophy and was unsure how to combine his interests. One day, Egami attended a workshop about U.S. graduate school, which he thought was about MBA programs. Actually, it was about PhD programs, and included a political scientist talking about using math in the field, so Egami asked her a question. 

“The miracle is: That workshop had 200 people in it, and after it was done, I was packing my stuff to go home, and the panelist, who was a PhD student, came down from the stage and found me,” Egami recalls. “She asked, ‘Are you the one who said you’re interested in political science in the U.S., and likes math?’” 

She invited Egami to what he thought would be another career workshop, the following week. Once again, he was mistaken.

“I showed up, and it was an academic seminar,” Egami continues. “There were only 20 people there. It was 19 professors, and me, a first-year undergrad.” Then a professor named Kosuke Imai, now at Harvard University, gave a talk about his own research on using statistics in the social sciences. 

“I was super-excited and felt if I could do even 20 percent of that, it would be a dream,” Egami says. “I talked to Kosuke and said, ‘I want to do what you’re doing.’ He probably thought I was just a random person.”

Egami, thus bolstered, started pursuing the goal of becoming a political scientist. He received his BA after spending a year as an exchange student at the University of Michigan, and applied to graduate schools in the U.S., landing at Princeton University — where Imai eventually became one of his advisors. Working with Imai, Rafaela Dancygier, Brandon Stewart, and others, Egami generated papers on methodological topics like external validity — and found substantial interest when he presented them. 

“That was a case where the audience or market told me what I should really work on,” Egami says. After earning his PhD from Princeton in 2020, he joined the faculty at Columbia University, moving to MIT five years later. 

Enjoying the spirit of MIT

One of the hallmarks of Egami’s work is very close scrutiny of the factors that can influence the results found in empirical studies. 

“In statistics, you talk about whether the people in the data are similar, meaning the population data,” Egami says. “But in political science, there are a lot of differences in context.” 

Consider the question of how much political campaigns sway the minds of voters. Political scientists have sometimes received permission to conduct field experiments in active political campaigns. That’s a significant step toward generating robust results. And yet, not all campaign settings are the same. Politicians may let researchers in when they expect to triumph, and the dynamics in those races might differ from close races. 

“It’s great to do field experiments, and that’s usually where people are allowed to do research,” Egami says. “It’s where politicians know they can win. But most of the time, we’re interested in the battlefield races, the politically competitive districts. And the logic and voter behaviors can be different in those cases.” 

Egami’s job, on one level, is to spot such differences and make other researchers aware of them.

Meanwhile, he has also developed a strong interest in scrutinizing the tools of machine learning, as applied to the social sciences. This predates the elevated interested in AI generated by ChatGPT, starting in late 2022. Some of Egami’s work explores how to systematically identify errors introduced by AI tools and then account for this issue when using AI in research.

“In the past, social science data is something we carefully collect and take a long time to really validate before we analyze it,” Egami says. “But if the generation of data is changing. If people use AI to generate data at scale, it can have errors. So I was already thinking: You want to have statistical methods that take into account these errors, otherwise many of the analyses will not be able to be replicated. That’s how I started to work on a lot of things about AI.” 

All of this has brought Egami recognition and honors in the field. Last year, he received the Emerging Scholar Award from the Society for Political Methodology. He has also been the recipient of best paper awards from the American Political Science Association’s sections for political methodology (in 2019 and 2025), experimental research (in 2024), and political networks (in 2022). Earning awards in three subfields of the discipline speaks to Egami’s scholarly versatility.

In his view, though, the work he does in different areas is ultimately aligned. 

“All these things are in parallel,” Egami says. “I’m trying to start a new research agenda every three to four years. That helps me learn new topics and be motivated.”

Further motivation, he says, comes from being at MIT and liking the experience.

“I already knew MIT was an amazing place I would enjoy,” Egami says. Even so, in his time at MIT, he says, he has gained even more appreciation for the “spirit of engineering,” in the sense of working systematically on solutions to ongoing problems, among other things. In any case, Egami has found the Institute to be a stimulating and congenial place to do his work. 

“People are really nice at MIT,” says Egami, who has been teaching both undergraduate and graduate classes.

He adds: “The Department of Political Science is really high-functioning, people are intensive in terms of their work, but it’s just genuinely nice people.” 

And, yes, that’s one claim about the world Egami does not have to double-check. 

At the beginning of their first year, many MIT undergraduates choose to join one of the Institute’s 44 fraternities, sororities, or independent living groups (FSILGs), some of which are housed across Cambridge, Boston, and Brookline, Massachusetts.

There are 30 fraternities, nine sororities, and five independent living groups for students to choose from. Nearly 37 percent of undergrads join an FSILG, and these communities offer students more than a place to live, eat, and socialize; they are places where students create friendships, mentor younger students, work with both alumni and MIT administrators, raise funds for local charities, and learn valuable leadership skills.

While each organization has its own set of values, traditions, and membership process, they all share a common goal: creating communities where students can grow both personally and professionally inside and outside of the classroom, while navigating the rigors of an MIT education.

Anya Kattef ’98, director of FSILG Alumni Programs, says, “I can't imagine my MIT experience — or the decades that followed — without the extraordinary community I found in Alpha Phi. Surrounded by smart, compassionate, and driven women, I gained the confidence not only to survive MIT's demanding academic environment, but also to grow as a leader, progressing through the officer roles of athletic chair, house manager, and ultimately president. Beyond the leadership opportunities, the mentorship I received from upperclassmen helped me secure my first summer internship, navigate course selection, and pursue opportunities I might otherwise have overlooked. And perhaps most meaningfully, the friendships I formed through Alpha Phi while at MIT have grown into lifelong bonds that continue to shape and enrich my life.”

Service is a common bond

Liz Jason, associate dean and director of FSILGs at MIT, says “although every organization is unique and has its own personality, service remains a common thread throughout every fraternity and sorority. Many national organizations partner with causes ranging from heart health research and children's hospitals to literacy initiatives. Local chapters then build additional partnerships with organizations throughout Greater Boston, supporting causes such as Rosie's Place, the Boston Area Rape Crisis Center, animal welfare organizations, and other community nonprofits.”

FSILGs often host signature fundraising events tied to philanthropy, while others organize volunteer opportunities throughout the year, such as cleaning up Back Bay alleys, so that it’s woven into the members' experience.

Jason also notes: “Our culturally based fraternities and sororities place a particularly strong emphasis on community service, with some requiring prospective members to demonstrate volunteer work before joining. In addition, some of our national organizations require students to complete at least one semester of college before joining to ensure they have established academic success first.”

Leadership and responsibility

Presidents and leaders of an FSILG take on a large amount of responsibility that goes beyond the scope of organizing social events or fundraisers. They’re managing organizations that function much like a small business.

“Leaders learn soft skills overseeing budgets, coordinating recruitment, mentoring new members, organizing educational programming, and often spend 10 or more hours each week fulfilling leadership responsibilities,” says Jason. “Leaders also learn conflict resolution while navigating disagreements among members or neighboring residents. They practice delegation, budgeting, prioritization, and time management. They gain experience running meetings, communicating with alumni volunteers, and working with senior Institute leaders. They have a seat at the decision-making table. As a leader, if you expect your peers to do something, you need to model and espouse that behavior, too.”

For students living in chapter houses, the responsibilities can extend even further. Leaders learn to manage multimillion-dollar properties. Student leaders coordinate building maintenance, communicate with vendors, oversee safety inspections, organize chores, and help maintain properties that, in some cases, have housed MIT students for more than a century. The student house manager manages the facility, attends training four times a year, where FSILG leadership goes over seasonal items they need to know, such as removing snow from sidewalks and steps, liability insurance, and safety inspections.

As the chapter president of Pi Beta Phi, senior Tea Picconatto says, “My role as president has strengthened my communication, leadership, and conflict-resolution skills. It has also connected me to the broader national organization and provided opportunities to build relationships with members and alumnae across the country. From a professional perspective, the experience has been valuable in demonstrating leadership and responsibility to future employers. I’m certain I was hired for two of my internship roles because of my sorority leadership experience.”

Picconatto adds, “Greek life offers a unique sense of identity, community, and connection to a nationwide network of members and alumnae that continues well after graduation in a way that is not replicated elsewhere on campus. My sorority sisters have always been there to offer emotional support, academic guidance, and encouragement whenever I have needed it.”

At MIT, Alpha Delta Phi Society is a gender-inclusive member of the Institute’s Interfraternity Council. As president, Gabriel Tian, who came to MIT from Toronto, Ontario, sought a community with which to experience MIT. During the first week of school, he was studying at the ADPhi house library late at night and said it felt very natural and productive. He says he thought “this is where I belong,” and pledged shortly after. Tian quickly became involved as academic chair and vice president, and even helped update the chapter's website.

“I have learned so much since Rush — how to be a leader, how to make difficult decisions, how to have hard and personal conversations, how to run a living community with an executive board, how to socialize more effectively to connect with each and every member. Being the president, or any other leadership position, is tough, but so incredibly valuable, and gives me confidence in myself and my ability to care of my community,” says Tian.

“In just two years since joining, I have made lifelong friends. In fact, some of the closest friendships in my life are right here in the siblinghood. The web of connections my chapter offers really enables connections between people who otherwise would not have the opportunity to have even met at MIT. To me, ADP makes MIT all the more brilliant and special.”

After graduating

Long after graduation, many alumni continue volunteering with their chapters, serving on house corporations, mentoring students, and helping preserve traditions for future generations. For many, the relationships formed during college continue throughout their professional and personal lives.

Some families even span multiple generations of FSILG life, with parents and children joining the same FSILG years apart. Others have found lifelong friendships — or even spouses — through their chapter experience.

Cecilia Warpinski Stuopis ’90, the chief health officer at MIT Health, found that when she joined the Alpha Chi Omega sorority while a student, she had an “instant group of peers.”

“I was trying out for the volleyball team, and my teammate invited me to Rush to see what it was about. We both were invited to join. There were about 20 of us in our pledge class, and perhaps 40 women total in the sorority at the time, and I’m still friends with many of my Alpha Chi sisters to this day. We’re a very tight-knit group. Sororities are a very supportive network of people who care about each other.”

Stuopis, whose husband also graduated from MIT, adds, “I became reengaged with the community at MIT as an alum, as a volunteer, and then an advisory board member for Alpha Chi Omega. My daughter came to MIT and pledged Alpha Chi, too, and this allowed me to attend her initiation. I’ve been on the Board of the Association of Independent Living Groups at MIT for the last nine years and recently signed up for another three. At MIT, fraternities or sororities are not like they are portrayed in the movies. They are guided by friendships, developing bonds, and are there to support all aspects of their member’s success — both during their time as students and well into the future.”

Students interested in joining an FSILG can find more information on the website.

In 2021, MIT set campus decarbonization goals as part of its Fast Forward climate action plan. Five years later, many of those goals have been met or are on track for completion, including efforts to make the Institute’s buildings more efficient, expand rooftop solar installations, and attain net-zero emissions. 

“Decarbonizing our campus goes hand-in-hand with MIT playing a leadership role in promoting carbon reduction and climate resilience through its research, innovation, and efforts to inform public policy in this area,” says Glen Shor, executive vice president and treasurer. “Our teams are leveraging that same innovative spirit to meet our campus climate goals.”

Creating an energy-efficient campus

Over the past decade, the Institute has decreased energy use per square foot by more than 10 percent, even as the campus has grown and research activity has intensified. Rooftop solar power generation has increased by more than five times in the same period, with installations added to the Stratton Student Center (Building W20), the Dewey Library (Building E53), the New Vassar undergraduate residence hall (Building W46), Graduate Junction (Buildings W87 and W88), and the theater arts building (Building W97). Thirty-three MIT building projects have earned Leadership in Energy and Environmental Design (LEED) certification. And in May, the Tina and Hamid Moghadam Building (Building 55) became MIT’s first Living Future Zero Carbon Certified building.

“We’ve completed more than 300 energy-efficiency projects across campus, focusing on our most energy-intensive research buildings, and ultimately touching nearly every corner of MIT,” notes Joe Higgins, vice president for campus services and stewardship.

Case in point: Building 46, home to the Brain and Cognitive Sciences Complex, and the Metropolitan Storage Warehouse (Building W41), newly home to the School of Architecture and Planning.

Building 46 was identified as one of MIT’s biggest energy users and the building with the greatest carbon-reduction potential. In 2024, the Institute completed a lab-by-lab renovation and improved Building 46’s mechanical systems infrastructure. The result: a 35 percent reduction in building energy use and carbon emissions — roughly a 2 percent reduction in overall campus emissions. 

The newly renovated Met Warehouse, which opened in August, features an innovative heat-recovery system, capturing heat rejected from the campus cooling system and using electric heat pumps to generate heat for the building. Higgins says the system will help inform the design of larger, campus-level heat-recovery systems.

Since 2014, 101 of the 168 buildings on MIT’s Main Campus have undergone energy-efficiency upgrades. The Institute’s 2030 Capital Plan will continue to invest in projects to reduce energy consumption and make efficiency upgrades a core element of all comprehensive building renewal projects. Examples of new projects include further optimizing heat-recovery systems; deploying more sophisticated controls to better manage ventilation, heating, and cooling; and using artificial intelligence to set classroom and office temperatures based on weather forecasts, occupancy patterns, and the forecasted carbon intensity of the regional power grid.

The renovation of Building 39, which is set to be home to a next-generation quantum research laboratory, will incorporate energy-saving features and technologies, including advanced insulation and windows, a smart ventilation system, LED lighting with automatic controls, and heat-recovery systems to maximize efficiency. These integrated systems are projected to dramatically reduce energy use and carbon emissions — cutting them by approximately 70–80 percent relative to the existing building baseline. 

Similarly, the McCormick Hall (Building W4) undergraduate residence hall renovation, which began this summer and is expected to be ready for students by the fall 2028 semester, will add high-performance windows, LED lighting, ventilation energy recovery, and low-flow plumbing fixtures. The project will replace gas cooktops with electric induction, use low-carbon flooring, improve stormwater management, and enhance the courtyard with native plantings, which require less water and maintenance while supporting local biodiversity.

On the path to net zero

MIT’s decarbonization efforts extend well beyond its campus. In recent years, the Institute has entered collaborations to create several large-scale renewable energy projects in regions of the United States where electric grids are still heavily reliant on fossil fuels. Together, these projects avoid over 200,000 tons of carbon dioxide per year, about equal to MIT’s annual direct campus emissions. 

“These projects, within a very short window of time, have had a significant impact on reducing emissions,” says Higgins. “They also put us on track to reach our net-zero target this year.” 

The first of these projects, the Summit Farms 60-megawatt solar farm in North Carolina, went online in 2016. Big Elm Solar in Texas, a 200 MW facility, followed in 2024, and Bowman Wind, a 208 MW wind farm in North Dakota, began operation in December 2025. Together, Big Elm and Bowman represent a landmark collaboration between MIT and 11 Massachusetts nonprofit and public sector organizations, including the City of Cambridge.  

“It’s a new market model that allows smaller organizations and government agencies to achieve greater reductions in carbon emissions that wouldn’t be possible on their own,” says Julie Newman, MIT director of sustainability. 

To capture the broader benefits of these projects, the Office of Sustainability worked with Institute researchers to develop a framework that assesses not only avoided emissions, but also economic and health outcomes. The team found that the projects generate economic benefits comparable to 7,000 one-year construction jobs and 189 maintenance jobs over 20 years. The projects’ annual health benefits are equivalent to 640 people quitting smoking for life, or nearly 200 premature deaths avoided each year for 20 years.

“Greener power sources are one of the building blocks we need to decarbonize our cities and campuses for the long run,” says Higgins. “That’s why we have made decarbonizing regional electricity grids a priority.”

The building blocks of campus decarbonization

To fully decarbonize MIT’s campus, the Institute will need to significantly change how it produces and distributes energy.

Currently, MIT’s Central Utilities Plant (CUP) burns natural gas to create electricity and steam-based heat, while also getting a small amount of electricity from the power grid. Electricity, heat, and air conditioning are distributed to campus buildings through a network of underground power lines and pipes. 

To move away from burning natural gas, and to take advantage of electricity from a greening grid for making heat, MIT is exploring creating a large-scale electric heat pump plant adjacent to the CUP on Vassar Street. The plant, a key building block for a long-term campus decarbonization strategy, will produce hot water and distribute it to campus buildings through a hot water-based heating system. 

“We’re starting the design process now, and in the coming year, we should know more about the scale and phasing of the heat pump plant we would construct, how it would interface with our existing district energy system, and the implementation timetable,” says Vasso Mathes, senior campus planner in the Office of Campus Planning, who is the campus decarbonization program manager. The heat pump plant will aim to recapture waste heat from existing cooling systems, supplying source energy to meet 30 to 40 percent of campus heating needs.

Another critical building block is transitioning MIT’s existing steam-based infrastructure to a hot-water system. That work — already underway — includes replacing steam distribution pipes to buildings with more efficient, easier-to-maintain hot-water pipes and converting buildings from steam to hot-water heat.

The third building block of a campus decarbonization strategy will be MIT’s ability to rely on the power grid for electricity instead of the CUP. “The electricity generated by the CUP is 15 to 20 percent lower in carbon emissions than the New England grid,” says Mathes. “We expect this to change over time as more and more renewables are added to the grid.” Even then, the CUP would be maintained as a backup system for use during peak heating and cooling days and grid stress events.

Finally, “the fourth building block is to go bigger, and look at shared infrastructure and coordinated planning with neighboring institutions and municipal partners,” says Higgins.

In that vein, earlier this year MIT became an anchor institution in the BosTEN Project, a year-long study to explore the feasibility of creating what could become the first city-scale thermal network in the United States. The network would help decrease the carbon footprints of major buildings across Boston and Cambridge, Massachusetts, by harnessing heat from the soil and rock under the Charles River and Boston Harbor, as well as waste heat from buildings and industrial facilities. It would also provide a renewable source of energy that can stabilize and even reduce the costs to heat and cool buildings.

“We’re thinking through how we can not only decarbonize our campus, but also how to use our work as a catalyst for broader strategies and technologies that others could readily employ,” Higgins says. “The unit of change needs to be at the city scale.”

For decades, one of the major complaints about nuclear power in the United States has been the argument that, after all this time, we still have not come up with a dependable strategy for sequestering high-level radioactive waste, including spent fuel from plant operation. This issue is of such importance that Haruko Wainwright has put it at the center of her research agenda as an Atlantic Richfield Career Development Professor in Energy Studies at MIT and an associate professor in the departments of Nuclear Science and Engineering and Civil and Environmental Engineering. 

In an essay called “The best-managed industrial waste in history,” which appeared in the Aug. 27 issue of the journal Nature, Wainwright made a bold statement, maintaining that an expansion of the nuclear power sector in the United States will benefit the environment, despite the fact that a solution to the permanent disposal of nuclear wastes has yet to be demonstrated in this country. 

In this interview, Wainwright describes risks associated with different forms of waste, ways to improve waste-handling procedures, and what lessons other countries can teach the U.S. in this realm.

Q: Why do you think chemical contaminants pose a greater public health risk than radioactive wastes?

A: I’ve always appreciated the fact that the dangers of radiation were recognized relatively early in the 20th century, prior to the widespread use of nuclear technologies. By the time an industry emerged, radiation protection standards were reasonably well established, including waste management. While nuclear power plants inevitably produce highly radioactive spent fuel, it is both solid and compact, making it relatively easy to contain and isolate from the environment. It took time to develop a disposal solution because people were pursuing a perfect one. Now, several countries are demonstrating that effective isolation over geological timescales is feasible. Finland, in fact, is about to open the world’s first deep geological repository for spent fuel.

Chemical contaminants present a different story. For many substances, like hexavalent chromium and PFAS (“forever chemicals”), the risks were identified long after they’d been released, having spread widely through the environment, food chains, and human bodies. PFAS, for example, has been used in industry and consumer products since the 1940s, yet the first federal drinking water standards were not adopted until 2024. Chemical hazardous wastes — including substances that degrade very slowly or not at all — are disposed of in the shallow subsurface without the requirement of long-term predictive assessments.

This is not to suggest that radioactive wastes are without risk. However, public perception is often disproportionately focused on — often hypothetical — nuclear hazards, while underestimating the dangers posed by chemical wastes. This misalignment actually has an adverse effect on the environment and public health. It leads to the misallocation of resources, diverting funding — including taxpayer dollars — away from worrisome contaminants whose environmental and public health consequences are already occurring. 

Q: How can we improve our procedures for storing spent fuel as more nuclear power plants come into operation around the world?

A: The nuclear industry is becoming increasingly proactive about waste management. Some companies, for example, now incorporate spent fuel storage capacity directly into their power plant designs, formulating plans that cover the entire operating period. Research on waste streams from advanced reactors — and even fusion reactors — is also growing. This approach of thinking about wastes before any are produced — what I call “design from the wastes up” — is critical for long-term sustainability.

Although further technical advances are surely needed, communication remains another area with significant room for improvement. Transparent monitoring programs and effective communications have been shown to build public confidence and provide assurance. Additionally, I believe we should place a greater focus on the inherent properties of radionuclides, including their risk pathways and mobility. Long-lived radionuclides are weakly radioactive and emit little or no penetrating radiation; their health risks are associated with ingestion or inhalation, analogous to chemical carcinogens. Most radionuclides, including plutonium, have low solubility and a high affinity for soil particles, limiting their mobility in the environment.

Current research on spent fuel storage has been devoted mainly to the integrity of the metal canisters used to contain spent fuel. Attention should also be directed toward developing predictive understanding of radionuclide transport and about geochemical barriers to the spread of radioactivity in the unlikely event of a containment breach. These approaches would exploit the natural immobility of radionuclides to afford additional layers of protection — in keeping with the nuclear industry’s recent embrace of passive safety features.

Q: How can the United States move toward the permanent disposal of nuclear wastes, and what can we learn from the European and Canadian examples? 

A: Many people tend to dwell on political and social issues, while the underlying science is frequently left out of the conversation. Fundamental questions — regarding the true dangers of radioactive materials and the feasibility of safe geological disposal — often go unanswered, leaving nuclear waste a vague, almost mythological threat, rather than a technical and engineering problem.

In fact, many people in geoscience believe that the failure of Yucca Mountain — the proposed geological repository for high-level radioactive wastes in the U.S. — stemmed from the fact that the site was chosen for political rather than scientific reasons. In 1987, Congress amended the Nuclear Waste Policy Act to confine site characterization to a single location, abandoning the original plan to screen multiple candidates. This top-down decision, widely dubbed the "Screw Nevada Bill," generated vehement local opposition. In addition, Yucca Mountain is the only proposed repository in the world situated above the groundwater table and within a zone of fractured igneous rock, where radionuclides are relatively mobile. Demonstrating its long-term safety is, consequently, much more difficult than for other proposed repositories.

Europe's approach to waste disposal offers a stark contrast. Switzerland, for example, identified a preferred site after a transparent, scientific evaluation of multiple candidates based on technical criteria, earning community acceptance as a result. Sweden and Finland built trust through decades of patient consultations with the public. And in Canada, more than 10 communities voluntarily expressed interest in hosting a repository before one favored site was ultimately selected.

Another underappreciated difference relates to how public concerns are handled. In the U.S., worries about radiation and radioactive waste have often been brushed aside by experts. In Europe, communication professionals and experts are trained to address every concern sincerely, offering understandable, science-based explanations. Discussing those concerns, moreover, can provide valuable opportunities to identify knowledge gaps and improve safety.

I believe that selecting a geologically sound site and communicating the science clearly — in terms that anyone can grasp — are the essential first steps toward achieving the permanent and safe disposal of nuclear waste.

Deep in the ocean, brown algae and diatoms produce a complex carbohydrate molecule called fucoidan, which helps form the algae's protective outer layer. The fucoidan molecule is very difficult for microbes to break down because its chemical structure may include dozens of different linkages and branching patterns that vary from one algae species to another. This resistance to decay is one reason why fucoidan matters; when microbes struggle to break it down, fucoidan can sink deep into the ocean, carrying carbon with it and potentially storing it for long periods. This could make fucoidan an important player in the ocean’s carbon cycle.

For many years, scientists knew of individual bacteria that could break down pieces of fucoidan. But one fundamental question remained unanswered: Could a microbial community break it down completely, and if so, how?

A new open-access study published in Nature, led by Andreas Sichert, a former MIT postdoc now at ETH Zurich, and Otto X. Cordero, associate professor of civil and environmental engineering at MIT, provides an answer.

"No single bacterium can finish the job," says Cordero. "Instead, fucoidan is degraded through teamwork. Different bacterial strains specialize in different parts of the molecule, and together, their combined efforts get the job done far more efficiently than any one organism could manage alone."

A puzzle with 453 pieces

In order to understand how fucoidan breaks down in nature, the research team enriched a fucoidan-degrading bacterial community from coastal seawater samples. What they found was staggering: more than 453 different genes, each responsible for making an enzyme that can act on fucoidan, spread across eight bacterial strains the researchers isolated. On their own, none of these strains could fully break down the molecule.

But when the researchers used a new, rapid mass-spectrometry method, they were able to observe how bacteria consumed individual sugar building blocks — and a clear pattern emerged. All of that genetic complexity could be reduced to two roles. Some bacterial strains specialized in degrading fucoidan's fucose-rich "backbone," while others specialized in removing its side branches, which contain less-common sugars such as xylose and galactose.

When strains playing both roles were combined, something noteworthy happened: degradation didn't simply add up. Instead, it became synergistic and exceeded what the bacteria's individual activities could predict. The more complementary the strains' preference for sugar were, the stronger the effect became. In some cases, the paired communities came close to completely degrading the complex polysaccharide.

"The breakdown of one of the ocean's most abundant carbon pools rests on a division of labor," says Cordero, "not between particular strains, but between functional roles."

Turning complexity into predictability

The most surprising result was that this division of labor made the system much more predictable than its underlying complexity indicated.

The researchers developed a simple model that sorted bacterial activity into two broad categories: fucose, and the rarer sugars found in fucoidan's side chains. They trained the model using data from small communities containing just one to three bacterial strains.

The simplified model was able to predict degradation in communities containing up to seven strains, and its predictions also generalized to nine structurally different fucoidans from other kinds of algae.

"A predictive understanding of a complex system need not come from characterizing each of its parts," adds Cordero, "but from finding the right simplification." The finding suggests that scientists may be able to predict how efficiently other complex, carbon-rich biological materials are broken down in nature, even when their exact chemistry and the enzymes involved are only partly understood.

The researchers also found that bacteria with complementary capabilities often occurred together in samples taken from the natural ocean, suggesting that the division of labor observed in the laboratory may also play a role in the ocean.

The consequences extend well beyond the field of microbiology. 

The researchers propose a concept they call "diversity-limited degradation," in which the absence of the right combination of complementary bacterial specialists allows fucoidan to persist for longer instead of being broken down. This concept may help explain why some algal carbon stays in the ocean for extended periods, contributing to long-term carbon storage.

For biotechnology, the takeaway is more straightforward. Instead of engineering a single "superbug" that can digest tough and complex biomass, a more promising approach may be to bring together teams of microbes that already specialize in complementary tasks. These teams could potentially be used to process brown algal biomass and other complex polysaccharides on a larger scale.

Looking ahead

The broader promise, though, may lie in the approach, rather than the molecule. If hundreds of uncharacterized enzymes can be reduced to two measurable traits, the same strategy might work for other biopolymers whose chemistry has so far resisted description — and, more generally, for predicting what microbial communities do. 

"Here was a system with hundreds of enzymes acting on a molecule we still can't fully describe, and it turned out to be far more tractable than anyone expected," says Cordero. "What we found is that there's a level of organization above the individual enzyme, corresponding to traits we can measure and plug into simple models that predict function from (genomic) composition. When biology looks intractable, it may be that we haven't found the right level of description yet."

One question the work leaves open is a fundamental one. Fucoidan is abundant, and has been for a very long time, so why has no bacterium evolved to eat it whole? The researchers suggest answers on two levels: constraints within sugar metabolism itself, and evolutionary dynamics in which complementary specialists are continually regenerated rather than merged into one.

"Really, this is a question about how life on Earth is organized," says Cordero. "Why are the biochemical functions that drive the planet's elemental cycles distributed across many organisms instead of concentrated in a few? Explaining that is, I think, one of the frontiers of the life sciences."

In addition to Cordero and Sichert, the research team included co-authors from ETH Zurich, the University of Vienna, and the Tata Institute of Fundamental Research.

The work was supported by Simons Foundation through the Principles of Microbial Ecosystems (PRIME) collaboration.

MIT researchers have developed a new technique that helps generative artificial intelligence models find solutions to high-stakes problems.

In these settings, a plausible answer is not enough: The output often must also satisfy nonnegotiable safety, physical, or task-specific requirements, known as hard constraints.

The researchers developed a method that helps generative models meet these strict requirements without sacrificing the quality of their outputs. 

The key to their technique is to give the model more freedom during the generation process and enforce hard constraints on the final output, rather than at every intermediate step. 

In experiments spanning robotics, control of physical processes, and computer vision, the new method consistently satisfied the required constraints while identifying better solutions than existing techniques. 

This adaptable, plug-and-play technique works at deployment time, so it can be applied to pretrained generative models without retraining them. It can make such models more useful in applications where safety rules, physical laws, or other strict requirements cannot be violated. 

“The promise of generative AI is its ability to explore a rich space of possibilities, but the real world places boundaries on which possibilities are acceptable. Our approach lets us preserve that generative power while enforcing the nonnegotiable requirements of high-stakes or safety-critical applications,” says Navid Azizan, the Alfred H. and Jean M. Hayes Career Development Associate Professor in the Department of Mechanical Engineering and the Institute for Data, Systems, and Society (IDSS), a principal investigator of the Laboratory for Information and Decision Systems (LIDS), and the senior author of a paper on this technique.

Azizan is joined on the paper by lead author Zeyang Li, a graduate student in mechanical engineering and LIDS; and Kaveh Alim, a graduate student in IDSS and LIDS. The research appears this week in the IEEE Transactions on Pattern Analysis and Machine Intelligence.

Freedom to explore

Pretrained generative AI models, such as diffusion models like Stable Diffusion and flow-matching models like FLUX, are now widely available. These powerful models learn to create new data by transforming random noise. Their availability has enabled people to adapt them to a wide range of applications.

These highly capable models excel at providing answers that come close to satisfying most queries, but in safety-critical applications like robot path planning on a crowded factory floor, an answer that is “nearly correct” may not be good enough. 

For instance, a “nearly correct” path from one machine to another might still result in the robot colliding with a human co-worker.

In such safety-critical applications, users often employ a technique called projection-based sampling, which repeatedly forces the model’s partial solutions, called intermediate samples, to satisfy strict requirements during the generation process.

But constraining the entire generation process can prevent the model from reaching a better final solution. These methods also typically focus only on satisfying the hard constraints, missing the opportunity to improve other qualities of the solution, like reducing the length of the robot’s trajectory.

“For constraint satisfaction, what ultimately matters is the model’s final output, since the internal process is discarded. By not requiring every intermediate step to satisfy the constraints, we give the model more freedom to find high-quality solutions that are still feasible in the end,” says Li. 

The researchers developed an algorithm called HardFlow that steers the sampling process so that the final output satisfies the user’s hard constraints without being overly restrictive and is of higher quality.

Subtle steering

HardFlow reformulates hard-constrained sampling as a trajectory-optimization problem, using tools from the field of optimal control. This enables the framework to steer the model’s sampling trajectory toward a goal, making subtle corrections along the way while enforcing hard constraints on the final output.

“Control theory gives us a powerful framework for formalizing the optimal way of making these corrections,” Azizan says.

But solving the trajectory-optimization problem around an enormous neural network was no easy task. The model may have hundreds of interconnected layers that process data.

To make the problem tractable, the researchers leveraged the structure of flow-matching models to decompose the problem into a sequence of smaller, single-step subproblems. They then applied systematic transformations and approximations to derive an efficient, scalable algorithm that still finds a feasible solution. 

“Essentially, we transformed the trajectory-optimization problem into something that preserves the key properties of the original problem, but can be solved very efficiently at deployment time,” Azizan adds.

Reformulating the task as an optimization problem allows HardFlow to incorporate additional goals that can improve the quality of the final answer. For instance, HardFlow could find a collision-free path for a robot that is also the shortest distance to its goal.

“Our framework can jointly handle both aspects, which helps it perform much better than existing methods,” says Li.

Across experiments in robotic manipulation, maze navigation, and text-guided image editing, HardFlow achieved perfect constraint satisfaction while consistently outperforming baseline methods on measures of solution quality. 

For example, it enabled a robotic manipulator to avoid collisions with obstacles while also finding the quickest path to the target object. Most other methods either resulted in collisions or found paths that took significantly more time.

In addition, HardFlow’s computation time was comparable to or lower than that of most competing methods.

In the future, the researchers could extend the framework to settings in which the AI model itself can also be updated, so that constraint satisfaction and sample quality can be improved in a more adaptive manner. 

The world needs more homes. The world also has too much plastic. Perhaps the only thing those two problems have in common is that they’re hard to solve.

Atlas Building Composites, a spinout of MIT, is on a mission to address both problems with a single solution. The company has developed an AI-powered robotic manufacturing platform capable of turning single-use plastics into durable building materials.

The company emerged from MIT HAUS, a research effort in the MIT Department of Mechanical Engineering that’s short for “Home Architecture for Universal Sustainability.” Atlas uses waterless plastic recycling and large-scale composite additive manufacturing technology to make parts like home foundations, decks, and trusses for walls, floors, and roofs.

“Our mission is to convert waste plastic pollution into durable composites to build 1 billion homes,” says Atlas chair and co-founder A.J. Perez ’13, MNG ’14, PhD ’23, who is also an MIT research scientist. “You can’t divorce these things from each other. We’re not here just to build homes, and we’re not here just to recycle plastic. The conventional way of building homes involves cutting down trees, mining, refining, and a bunch of other dirty activities. We want to avoid all that and address all the plastic bound for our oceans and landfills. We’re turning bottles into buildings.”

Atlas’ parts are already being used to support barns, sheds, decks, and docks. Most recently, the company supplied the U.S. Army Corps of Engineers with American-made recycled composite trusses to construct a 40-foot bridge in a Massachusetts wetland.

Perez and Atlas co-founder Matt Pouliot envision deploying thousands of their AI robotic production systems around the world. A key enabler for that scale is the company’s ability to recycle low-grade plastic into building components without water.

“This is key to democratizing recycling,” Perez says. “Now, every country around the world, regardless of their water access, will be able to do something about their plastic. We strive to study these issues in the real world, not just a lab. When you talk to government officials about creating a new recycling facility, they have to get the local water agency involved, there’s permitting, etc. A lot of that work disappears with the waterless recycling process.”

Research for impact

Since earning his PhD at MIT, Perez has been developing advanced fabrication techniques for homes and new techniques for plastic recycling. In 2019, he started MIT HAUS with David Hardt, MIT’s Ralph E. and Eloise F. Cross Professor in Manufacturing.

“It started with the simple mission of enabling the production of 1 billion homes over a 30-year period,” Perez says. “Then we realized how much the materials needed for those homes would strain global supply chains.”

Perez says building those homes using conventional methods would require a doubling of global production capacity for materials like concrete, not to mention a dramatic acceleration of global deforestation.

“That’s where the light bulb went off,” Perez says. “There’s this other problem humanity has, which is 8 gigatons of plastic that have been produced and are polluting our oceans, rivers, and cities. We decided to plug two really big, hairy problems together.”

Perez met Pouliot, a former Maine senator, and the pair started Atlas to commercialize the technology Perez had been developing at MIT. The founders worked with MIT’s Technology Licensing Office and have since worked with researchers at other universities to independently develop technology for the company’s robotic manufacturing platform, which the founders call the Atlas Factory Stack.

First, single-use plastic from water bottles and other objects is shredded and fed into the Atlas system, where it is melted and fused with American-made fiberglass to make it stronger than wood. From there, a large-scale 3D printer creates the parts, including trusses for floors, walls, roofs, and bridges.

Through research at MIT, Perez has shown large composite trusses can be printed in under 13 minutes and support over 4,000 pounds, exceeding key building standards.

“At MIT, we’ve demonstrated we can produce 60 to 80 pounds of parts per hour, and the systems we’re specifying in Atlas factories operate in the 150 to 200 pound per hour range,” Perez says. “There’s the potential for our robotic manufacturing platform to produce each part at a lower cost than injection molding, and it’s far more flexible and convenient. For example,  we can manufacture the parts in the reverse order so that they’ll be placed on the finished goods pallet next to the machine.”

The founders envision Atlas as a technology provider enabling the creation of home factories close to wherever homes need to be built. Today, each Atlas factory cell is capable of producing the structural framing components for about one small home per day.

“The old way of doing things would be some huge factory in China would mass produce one type of part and ship it far away,” Perez says. “I don’t think that’s good for the planet. Another reason we don’t use injection molding is economic: Mega factories don’t produce as many jobs and have a much higher carbon footprint. We want this to be localized to benefit local communities. The plastic is already everywhere. The more local Atlas is, the lower the cost and footprint.”

Going global

Plastics last far longer than wood, especially for applications where they’re in contact with the ground or water. That adds to the company’s environmental benefits.

“If you get a material into the building world and it does its job, it’s going to be used for a very long time and not need to be recycled again for a very long time,” Pouliot says. “That’s important because when you recycle something over and over again, it degrades. This is one of the most sustainable use cases for recycled petrochemical products.”

Atlas’ bridge with the Army Corps of Engineers was installed in less than a day. The founders are also in talks with international franchise partners to deploy the Atlas Factory Stack across the globe.

“To accomplish our mission, I fundamentally believe it’s not going to be one far-away company dominating the industry,” Perez says. “It’s going to be every country leveraging Atlas Factory Stacks to create local recycling jobs, local factory jobs, local construction jobs, and to stimulate their economies with local materials.”

The Federal Laboratory Consortium (FLC) selected AI-GUIDE, a medical device developed by MIT Lincoln Laboratory and Massachusetts General Hospital (MGH), for its 2026 Excellence in Technology Transfer Award. This award recognizes federal laboratories and collaborators who have accomplished outstanding work in the process of transferring technology. With funding from the U.S. Army's Combat Casualty Care Research Program (CCC), Lincoln Laboratory and MGH developed AI-GUIDE and are in the process of transferring the prototype to the startup company AutonomUS Medical Technologies, Inc.

"This recognition reflects what effective technology transfer looks like — aligning the Army's operational need with Mass General's clinical expertise and Lincoln Laboratory's engineering capabilities to deliver a solution with a clear path to impact. The transition to AutonomUS underscores how strong partnerships can carry a technology from development into real-world adoption," says Asha Rajagopal, Lincoln Laboratory's chief technology transfer officer. 

AI-GUIDE's transition to industry promises improved health outcomes for injured service members and civilians. Unlike ultrasound devices typically found in hospitals, AI-GUIDE is small and portable, making it ideal for use in pre-hospital settings. Pairing custom-developed AI software with commercial handheld ultrasound technology, AI-GUIDE helps the user insert a guidewire and catheter into a patient's blood vessel. This capability is especially important for U.S. military medics, who must keep injured soldiers alive in the field — sometimes for days — before they can be evacuated to a hospital. AI-GUIDE allows medics with minimal specialized training to administer medical interventions that would otherwise be impossible outside of the hospital, drastically improving patients’ chances of survival. 

The AI-GUIDE project has served as a framework for effective technology development and transfer. Within just three years, AI-GUIDE went from an idea proposed by CCC to a fully working proof-of-concept technology with its own startup company. Once the prototype was developed, clinical testing at MGH proved its viability, and Lincoln Laboratory and MGH staff then founded AutonomUS Medical Technologies to facilitate the commercialization process. With support from the MIT Technology Licensing Office, Lincoln Laboratory Technology Transfer Office, and CCC, the company secured U.S. Food and Drug Administration (FDA) Breakthrough Device Designation, a regulatory fast-track pathway that is only granted to highly innovative technologies with lifesaving potential, as well as a Small Business Innovation Research grant from the U.S. Department of the Air Force and funding from private investors, the Department of War, and the National Institutes of Health. 

These strong technology transfer collaborations are designed to streamline the transfer process, ensuring that lifesaving capabilities can be made available to military personnel and civilians as quickly as possible. While much of the initial work on vascular access has already been transferred, the AI-GUIDE team continues to develop and transition additional capabilities, including peripheral nerve block technology for trauma care and pain management. AI-GUIDE has previously been recognized with a Lincoln Laboratory Best Invention Award and an R&D 100 Award. 

"Lincoln Laboratory has a long record of transferring technology to industry. We are honored and proud to be recognized for the transfer of AI‑GUIDE and look forward to seeing the technology commercialized and saving lives in the field. This achievement reflects the strength of the partnership among the Defense Health Agency, Lincoln Laboratory, Massachusetts General Hospital, and AutonomUS Medical Technologies," says Samuel Kesner, a technical staff member in the Systems Engineering Group, who currently oversees the AI-GUIDE program at Lincoln Laboratory. 

Winning team members from the laboratory include Brian Telfer, Samuel Kesner, Lars Gjesteby, Joshua Werblin, Benjamin Roop, Alec Carruthers, Nancy DeLosa, and former Lincoln Laboratory staff members Matt Johnson (now the vice president of engineering at AutonomUS) and Laura Brattain (now an associate professor at the University of Central Florida). Asha Rajagopal, Jordan Mizerak, Melly Coronado, and Jonathan Dan supported technology transfer efforts.

This summer, the MIT Schwarzman College of Computing welcomed faculty from colleges and universities across Greater Boston, South Carolina, West Virginia, and Texas to campus for the inaugural AI Educators Pilot, a weeklong workshop aimed at expanding how artificial intelligence is taught across disciplines and learning environments. 

Inspired by MIT class C01/C51 (Modeling with Machine Learning), a course developed through the Common Ground for computing and AI education that focuses on helping students understand and apply foundational AI and machine learning concepts to problem-solving in their own disciplines, the workshop gave educators an opportunity to explore how its materials and teaching methods could be adapted for their classrooms. 

“The broader goal is to expand AI education to more students by investing in training for instructors,” says Dan Huttenlocher, dean of the MIT Schwarzman College of Computing and the Panasonic Professor of Electrical Engineering and Computer Science (EECS).

“We want to empower students to become critical thinkers about AI, not just users of the technology,” says Asu Ozdaglar, deputy dean of academics for the MIT Schwarzman College and department head of EECS.

A collaborative model for expanding AI education

Bringing the program to life required broad collaboration across the college, including support from leadership, staff, and contributions from more than half a dozen instructors in fields ranging from finance and computer science to sustainability. Together, they helped shape a workshop that paired core technical concepts with examples and teaching materials adaptable to a range of classroom settings.

“I have not seen an effort quite like it — this many dedicated instructors assembling materials of this richness, all to equip the educators who serve their students,” says Saurabh Amin, the Edmund K. Turner Professor in Civil Engineering and faculty director of the AI Educators Pilot. Amin is also co-director of the Operations Research Center, which is jointly housed within the MIT Schwarzman College of Computing and MIT Sloan School of Management.

With support provided by Jake and Robin Reynolds, the pilot brought together 19 participants in July from Allen University, Babson College, Brandeis University, Marshall University, the University of Massachusetts at Lowell, the University of North Texas, and Wentworth Institute of Technology. Working alongside MIT faculty and instructors, participants explored the pedagogy behind Modeling with Machine Learning through a mix of demos, videos, and exercises, and collaborated in hands-on activities focused on translating the course’s materials and methods to their own classrooms.

“This opportunity has been very timely because we are starting an AI and data science program in my department,” says Wenjin Zhou, assistant professor of computer science at UMass Lowell. “We’ve already been thinking about: How do we teach our next generation of computer scientists within the area of AI? How do we integrate AI in the teaching? I wanted to learn more about how other people are doing it, and especially answer the question: If AI can create tools for anyone now, what does a computer scientist do?”

Moving beyond the black box

When it comes to AI, Amin notes, there is no shortage of high-quality material. What is usually missing is context: Opportunities for instructors and students to connect AI concepts to specific disciplines, problems, and ways of thinking. Those connections are often built through dialogue and reasoning, rather than by presenting AI as a fixed set of ideas to be received. But instructor capacity remains one of the scarcest resources.

“What is scarce are educators prepared to teach AI as more than a fixed body of concepts and tools, to ground it in their own field, help students use it with judgment, and demystify it, so students do not just apply models but learn to question, adapt, and build with them,” explains Amin.

Shen Shen, an EECS lecturer and one of the workshop instructors, adds, “How do we make sure that machine learning is not just a black box, nor this magic piece of new technology? You can think of it as a tool, or a new framing to help you solve the problem in your specific domain.”

From pilot workshop to educator network

Participants ended the week by reflecting on which workshop materials and teaching approaches they planned to adapt for their disciplines and courses. Their feedback will help shape future iterations of the pilot and support the development of a broader network of educators committed to expanding AI education across diverse learning environments.

Weijie Pang, an assistant professor of computer science at the Wentworth Institute of Technology who attended the workshop, looks most forward to ongoing community building activities. “This is a really valuable opportunity to communicate with other faculty from different majors and areas. I can see what other universities are doing and what we can learn from each other,” she says.

“It's helpful to know that everybody within different disciplines at different universities is struggling with the same questions of how we can best serve our students as the technology is changing. Hopefully, we can set them up for success by being a little bit more forward and anticipatory of what the AI use is going to be,” says Dylan Cashman, an assistant professor of computer science at Brandeis University.

The brain cancer glioblastoma is one of the most aggressive and treatment-resistant cancers known to medicine, carrying a median survival of just 12-15 months, even with the best available care. Now, researchers at the MIT Media Lab have developed injectable nanoantennas, each about one-hundredth the width of human hair, that can be magnetically activated to create localized therapeutic electric fields that target and kill brain cancer cells without damaging healthy brain tissue.

“In laboratory and animal studies, this approach significantly reduced tumor growth and extended survival without detectable side effects, highlighting its potential as a precise and safe brain cancer therapy,” says Deblina Sarkar, associate professor and AT&T Career Development Chair at the MIT Media Lab and head of the Nano-Cybernetic Biotrek group.

The researchers named their technology “HITMAN” — short for highly-localized electric-field-induced tumor therapy using magnetically actuated nanoantennas. 

An open-access paper describing this technology published today in Science Advances.

To test HITMAN against the most clinically realistic version of this disease, the research team worked with tumor tissue obtained from patients diagnosed with aggressive and chemotherapy-resistant glioblastoma at Mayo Clinic. Using cells derived from this tissue in the laboratory, the researchers demonstrated that HITMAN eliminated 52.2 percent of these drug-resistant cancer cells — more than five times than that achieved by the standard chemotherapy drug temozolomide (TMZ) — while leaving healthy neurons and brain-supporting astrocytes unharmed.

The team then implanted those patient-derived tumor cells into the brains of mice to recreate the disease in a living system. In these orthotopic animal models — widely regarded as the gold standard for preclinical brain tumor research — HITMAN substantially inhibited tumor growth, extending median survival by more than 50 percent with no detectable toxicity to major organs or surrounding healthy tissue. 

The injectable nanoantennas can be activated wirelessly from outside the body, with the application of a low-frequency (no higher than 200 kHz, to prevent tissue-damaging heat) magnetic field that can penetrate the skull and brain tissue. The magnetic field actuates parts within the nanoantennas made of magnetostrictive material, creating stress and strain, which result in deformation of a piezoelectric film, producing localized electric fields.

Such localized electric fields were demonstrated to preferentially attack glioblastoma at the cellular level, disrupting the cells’ inherent bioelectric currents and fields, which regulate cellular function. Such disruption provoked a number of antitumor mechanisms, including protein unfolding, membrane damage, and endoplasmic reticulum stress, curtailing the production of a cell’s functional proteins. Such forms of cell dysfunction led to cell death. According to the researchers, cancer cells were selectively targeted over healthy cells due to their high proliferative rate, which elevates protein-folding demand, as well as their characteristic abnormalities in membrane composition and intracellular organelles.

Among a wide array of control experiments, the researchers also exposed glioblastoma cells to the nanoantennas without applying a magnetic field, as well as exposing the cancer cells to a magnetic field alone, confirming that the demonstrated effects were in fact due to the nanoantennas and their magnetic field activation. They also tested for side effects damaging to the animal models’ major organs — kidneys, liver, spleen, lungs, and heart — and detected none.

Also demonstrated by the research was a significant reduction in the number of cancer cell colonies formed after application of the nanoantennas, from 112-150 in the control groups to just 26 in the experimental group, indicating significant potential to reduce tumor recurrence and metastasis.

If translated to clinical use, the nanoantennas, whose size is approximately 150 nanometers, could be injected through the skull. Sarkar points out, however, that a technology developed previously in her lab could make their deployment even simpler.

In 2025, Sarkar and her colleagues created “circulatronics,” a technology that could allow devices like the HITMAN nanoantennas to be administered through an injection in a patient’s arm and to travel to a target region of the brain. In that previous work, the electronic devices were integrated with living cells so they would not be attacked by the body’s immune system and could easily cross the blood-brain barrier, as was demonstrated in pre-clinical studies. 

A glioblastoma diagnosis comes with formidable treatment challenges. Because this type of cancer is extremely infiltrative, complete tumor removal is difficult to achieve and can affect cognitive function. Also, the tumors often resist radiotherapy and chemotherapy, and immunotherapy is challenged by an immunosuppressive tumor environment. 

“The persistent failure of these therapies underscores the urgent need for novel approaches to target treatment-resistant glioblastoma cells,” the researchers write. “HITMAN offers a minimally invasive, spatially precise, and clinically translatable therapy for glioblastoma.”

Sarkar is joined on the paper by other members of her lab, including Monochura Saha, a former MIT postdoc; Ishaq Khan, a former MIT senior postdoc; Baju Joy, Shun Ying Chen, Hao-Tung Yang, Preet Patel, and Pengrui Zhang, all MIT graduate students; and Faheem Azeemi, an MIT undergraduate student.

During the day, waste products such as lactic acid and worn-out proteins build up in the brain. When we sleep at night, waves of cerebrospinal fluid (CSF) help to wash away this waste, keeping the brain healthy.

In a new study, MIT researchers have shown that they can strengthen these CSF waves through exposure to short bursts of a gentle, staticky sound known as “pink noise” during sleep. These bursts increase the amplitude of slow electrical waves in the brain, which then enlarges the CSF waves.

The researchers now hope to explore whether this enhanced CSF flow could help to boost cognitive function, improve memory, or even slow the progression of neurodegenerative diseases caused by the buildup of harmful proteins such as amyloid beta.

“We found that we were able to increase the size of the CSF flow wave during sleep, which as far as we know, there hasn’t been a method to do before. Now that we can enhance CSF flow during sleep in healthy adults, we’re really excited to bring this technology to clinical populations to see what effects we can have,” says Laura Lewis, the Athinoula A. Martinos Associate Professor of Electrical Engineering and Computer Science, a member of MIT’s Institute for Medical Engineering and Science and the Research Laboratory of Electronics, and an associate member of the Picower Institute for Learning and Memory. 

Lewis is the senior author of the study, which appears today in Science Translational Medicine. Joshua Levitt, who recently earned his PhD from Boston University and was a visiting graduate student in Lewis’ lab, is the paper’s lead author.

Cleaning up the brain

Cerebrospinal fluid is a clear liquid that surrounds and cushions the brain and spinal cord. In addition to protecting the brain from injury, it also helps provide nutrients such as glucose and removes waste products secreted by brain cells as they burn energy.

In 2019, Lewis reported a way to use functional magnetic resonance imaging (fMRI) to measure CSF waves as they flow in and out of the brain during sleep. That study showed that these waves are tightly coupled with brain waves called slow waves, which are associated with deep sleep.

In the new study, she wanted to further explore the relationship between brain waves and CSF flow, and investigate whether manipulating brain waves might enhance CSF flow. Previous work had already shown that delivering an auditory stimulus at the peak of slow waves can deepen the waves.

“You can make more of these electrical slow waves through an auditory stimulus, if it comes at just the right time. Similar to a child on a swing, if you push them when they’re at the right moment in their movement, you can make that swing go farther,” Lewis says. “The challenge is: How do you find just the right time?”

The auditory stimulus used for this study is a 50-millisecond burst of pink noise. Similar to white noise, pink noise contains all sound frequencies audible to the human ear, but the lower pitch frequencies are louder and the higher pitch frequencies are softer. This creates a balanced, gentle sound similar to steady rain or a distant waterfall.

To deliver these bursts at the peak of the brain’s slow waves, the researchers had to measure each participant’s EEG activity as they slept. This proved challenging because they also needed to measure fMRI signals to monitor CSF flow, and the magnetic fields used for fMRI interfere with EEG signals.

To overcome that, the researchers developed a way to process the EEG signals to eliminate the noise caused by fMRI, very rapidly — in less than 100 milliseconds. To make up for that small lag time in the EEG measurement, they also developed an algorithm that could predict when the slow wave peaks would occur. This allowed them to deliver the pink noise stimulus at the correct time.

More restorative sleep

In tests of 14 healthy volunteers, the researchers found that the auditory stimulus they delivered — which is not loud enough to wake a sleeping person — increased the amplitude of both the slow electrical waves and the CSF waves, during sleep.

Their fMRI studies also revealed that the slow waves stimulate blood vessels to constrict and dilate, allowing them to act as a pump that drives CSF out of the brain. Slow waves are seen only during non-REM sleep, and they become more prominent in deeper stages of sleep.

The researchers now hope to study whether enhancing CSF flow could help people to get more restorative sleep, especially people with insomnia. They also plan to explore whether increasing the flow of CSF, and the removal of waste products from the brain, could help people with Alzheimer’s and other diseases characterized by buildup of harmful proteins. 

“Brain waste clearance is really important for Alzheimer’s and other forms of dementia, which are caused, in part, by the buildup of molecules like amyloid and tau in the brain. If we can improve brain waste clearance, we may be able to help prevent the buildups of these plaques that lead to disease,” Levitt says.

Levitt has started a company that hopes to develop a device, such as a headband, that people could use at home to increase CSF flow by delivering an auditory stimulus at the right time. 

The research was funded by a McKnight Scholar Award, a Sloan Fellowship, a Pew Biomedical Scholars Award, the Simons Foundation Collaboration on Plasticity in the Aging Brain, the MIT EECS Transformative Research Fund, the National Institutes of Health, the Corundum Convergence Institute, and the Panasonic Well Fellowship for AI and Wellness.

As a liquid that is easily stored and transported, ammonia (NH3) is an attractive carrier for hydrogen, which is used in fuel cells, semiconductor manufacturing, chemical processing, and other applications. However, breaking ammonia into hydrogen and nitrogen typically requires high temperatures, and the resulting gas mixture must undergo additional purification before the hydrogen can be used in many applications. 

MIT researchers have now developed an electrochemical approach to promote hydrogen release from ammonia while simultaneously separating and concentrating the hydrogen into a high-purity stream. Their strategy, which uses electricity to speed up the extraction, reduces the temperature and energy required to recover hydrogen from ammonia and other hydrogen carriers.

In a new study, the researchers showed that their approach can generate highly concentrated, pure streams of hydrogen.

“We have shown the ability to use electrochemistry to drive thermodynamically uphill and kinetically difficult dehydrogenation reactions,” says Yogesh Surendranath, the Donner Professor of Science and a professor of chemistry and chemical engineering. “In this case, we studied the conversion of ammonia and a liquid organic molecule because of their importance as possible hydrogen carriers for a hydrogen economy. But the concepts we learned here could in principle be translated further, and we’re actively working on translating it to other important dehydrogenation reactions.”

Surendranath is the corresponding author of the study, which appears today in Nature. MIT postdoc Rui Zeng, now a professor of materials science and engineering at Harbin Institute of Technology in Shenzhen, China, is the paper’s lead author.

Extracting hydrogen

Hydrogen is widely used in semiconductor manufacturing and chemical processing and is also an energy carrier in fuel cells that use hydrogen and oxygen to generate electricity without combustion. Expanding its use, however, will require practical ways to store and distribute it.

Hydrogen gas itself is difficult to transport efficiently without compression or liquefaction. One alternative is to store hydrogen chemically in compounds that are liquids or can be readily liquefied, then release it where and when it is needed.

Ammonia is one promising hydrogen carrier because it is already produced and transported across large distances, but recovering hydrogen from ammonia remains challenging. That process, known as “cracking,” requires temperatures higher than 500 degrees Celsius to achieve high reaction rates and conversion. The hydrogen must then be separated from nitrogen and unreacted ammonia.

“We wanted to ask whether we could use electrical inputs to drive what would otherwise be an unfavorable dehydrogenation reaction, and simultaneously do it in a way that would separate the hydrogen from the hydrogen carrier, so that it would be very pure and could be used directly in a fuel cell or other application that requires a high purity hydrogen stream,” Surendranath says. 

The key element of the researchers’ new design is the coupling of a palladium-based separation membrane with a hydrogen-generating electrode through a molten hydroxide electrolyte. The separation membrane selectively transports hydrogen while preventing other components of the reaction mixture from passing through.

Using the new setup, ammonia is first dehydrogenated by a catalyst containing ruthenium and cesium. The hydrogen then reaches the separation membrane, whose opposite side is in contact with a molten hydroxide electrolyte. 

The electrochemical gradient across this membrane effectively creates a “vacuum” for hydrogen, providing a strong driving force for its transport across the membrane. It also converts the hydrogen into protons and electrons, which travel separately through the molten electrolyte and external circuit, respectively, before recombining at a second electrode to form hydrogen gas. 

Because the membrane selectively transports hydrogen, the system produces a concentrated stream of hydrogen gas without requiring a separate downstream purification process.

“Using this electrochemical process, we’re able to do this active pumping of hydrogen from a low concentration to a high concentration,” Surendranath says.

Continuously extracting hydrogen can also help drive the dehydrogenation reaction forward, especially when the presence of hydrogen inhibits the reaction. In this way, this strategy does more than separate the product: It changes the reaction environment and enables hydrogen recovery under milder conditions.

This process thus can be performed at temperatures around 200 or 300 degrees Celsius, much lower than those required for conventional ammonia cracking. Another advantage is that it creates a pure stream of hydrogen that doesn’t need to be purified later on — a step that requires additional energy.

Curtis Berlinguette, a professor of chemistry and chemical and biological engineering at the University of British Columbia, described the method as “a powerful new way” to solve the problem of obtaining a pure stream of hydrogen from ammonia and other hydrogen carriers. 

“By using electricity to pull hydrogen through the membrane as it is released, they accelerate the dehydrogenation of ammonia and liquid organic hydrogen carriers while simultaneously producing a purified hydrogen stream. This is an important advance for the energy sciences because it opens a credible pathway for transporting hydrogen in stable chemical carriers and releasing it where and when it is needed,” says Berlinguette, who was not involved in the research.

Powering transportation

In this study, the researchers showed that this approach could be used to dehydrogenate not only ammonia but also methylcyclohexane. This molecule is part of a class known as liquid organic hydrogen carriers (LOHCs), which also hold potential as an energy carrier.

The researchers envision that their new strategy could be useful for transportation applications, such as powering cars, buses, or ships, or for fabricating semiconductors or electronics. Pure hydrogen gas is used for several steps in semiconductor manufacturing, where it plays important roles in boosting manufacturing yields and reducing surface defects.

Because palladium is an expensive metal, the researchers are now working on ways to reduce the amount of palladium needed for the separation membrane. They are also working on scaling up the process, and on applying it to other dehydrogenation reactions that could be industrially useful.

The research was funded by the U.S. National Science Foundation.

How will global access to food, water, and energy evolve in coming decades? A new study co-authored by MIT researchers suggests the answers could be very different depending on region, resource, and income.

Based on extensive modeling of many different resource scenarios, the study finds that in some regions, lower-income people could be spending roughly 50 percent of their income on food by the year 2050, in contrast to higher-income groups that could spent about 5 percent of income on food in the same areas. 

“For a lot of these outcomes, the lower-income groups see much worse potential insecurity,” says Jennifer Morris, a principal research scientist at the MIT Center for Sustainability Science and Strategy and the MIT Energy Initiative, and co-author of a new paper detailing the findings. The results, she notes, can be evaluated by policymakers in different global regions to understand what the long-term, large-scale resource security risks may become for different parts of their populations. 

“Anything that’s taking up half of your income is potentially destabilizing for your entire life because it leaves so few resources for the other critical needs and basic life necessities,” Morris says. 

The study focuses on projecting future access to food, water, and energy, based on long-term variation across a dozen major factors influencing their availability, from economic conditions and agriculture production to trade conditions, climate, land use, and more. 

“This study shows that there is no single driver of future food, energy, and water insecurity,” says Gi Joo Kim, a research scientist at Tulane University and co-author of the paper. “Income is important, but regional conditions, land use, energy systems, water availability, and consumer behavior all shape the risks people face.” For policymakers, he adds, “This means they need to consider specific combinations of factors that create vulnerability in each region.”

The paper, “Identifying Key Uncertainties and Drivers of Future Resource Security Outcomes Through a Multisector Scenario Ensemble,” appears in the journal Earth’s Future.

In addition to Morris and Kim, the authors include Brian O’Neill, an earth scientist at the Pacific Northwest National Laboratory; Marshall Wise, a system engineer at the Pacific Northwest National Laboratory; John Weyant, a professor of management science and engineering at Stanford University; and Jonathan Lamontagne, an associate professor of civil and environmental engineering at Tufts University. 

Filling a gap

The current study fills a gap in modeling among scientists studying issues such as long-term resource security. Given the complications of long-term analyses, many studies have used what scientists term “shared socioeconomic pathway” circumstances, a small set of senarios spanning broad global narratives about the future, rather than exploring specific outcomes such as how long-term resource access may shift in linked fashion across income groups in different regions of the world. Two years ago, the same group of authors wrote a paper calling for more socioeconomically specific scenario analysis focused on outcomes for human well-being; the current study is their effort to develop that kind of modeling. 

“For this type of study, where we’re focused on human well-being outcomes, the income piece is really important,” Morris says. 

To conduct the study, the researchers adopted an existing framework in the field, the Global Change Analysis Model (GCAM) version 7.1, which represents interactions between energy, economies, water, land, and climate while dividing the world into 32 regions, 235 water basins, and 384 land-use regions and making adjustments for things like estimated commodity prices over time.

The research group used 12 main variables connected to resource availability, including population, GDP, income distribution, carbon intensity, land use, agricultural trade, multiple energy consumption scenarios, multiple water-use projections, and more. They ran simulations for 3,735 different scenarios involving these factors, to better understand the range of possible resource outcomes by 2050. 

Broadly, the modeling does uncover some significant regional variations. In 2050 food security may be most acute in parts of sub-Saharan Africa, while energy security could be most acute for low-income residents in some parts of Asia, Eastern Europe, and the Middle East. 

But within any region, there may still be substantial variation in resource security. In southern Africa, the modeling suggests that the poorest 10 percent of the population by income could be spending 49.6 percent of its income on food, compared to just 5.5 percent for the wealthiest 10 percent of the population. In West and East Africa the projected food burden for the bottom 10 percent of the population in terms of income is projected to be 48.4 percent and 42.5 percent, respectively. 

To understand the potential change this represents over time, the researchers compared the results to data from the year 2015 in the GCAM model. For the lowest-income group across western Africa in 2015, the average food burden was about 25 percent of people’s income, compared to estimates for 2050 that range from about 20 percent to 75 percent of income. In southern Africa, the lowest-income group spent about 20 percent of their income on food in 2015, but the scholars’ modeling projects an increase in food burden ranging from 25 percent to 65 percent of income. The wide variation in projected burden reflects the wide range in possible future scenarios.

When it comes to energy, variation by income is also apparent. In some parts of the Middle East, for instance, the residential energy burden in 2050 is estimated to be just 1.7 percent for the highest income bracket but 18.9 for the lowest income bracket; in Eastern Europe, the energy burden reaches 11.3 percent of income for the lowest-income bracket, while resting at under 5 percent for the highest-income bracket. 

“Regional averages can make future resource-security risks appear more manageable than they actually are,” Kim says. “This means analyses that stop at the average may miss exactly the populations most vulnerable to future change.”

Understanding the dynamics

To be sure, as the scholars emphasize, there are many uncertainties when it comes to resource access, and uncertainty is always part of modeling the global economy and resources. Still, they believe these kinds of projections can provide a more detailed outlook about social conditions in 2050 than has previously been available.

“At the very least, it’s highlighting areas of concern and showing that they differ in different parts of the world,” Morris says. “One of the outputs of this type of study is to map that out and provide that kind of insight. That can also inform the focus of further studies into specific regions and concerns.”

The researchers also believe the results will provide a new roadmap for policymakers who may be concerned about long-term resource provision across the entirety of their societies. While having new projections is valuable, modeling also helps analysts and policymakers see which factors most clearly influence future resource outcomes, as well.

“Our method was designed to identify the conditions that produce different resource security outcomes, rather than to predict one most likely future,” Kim says. 

“It’s a different approach to scenarios than we typically see,” Morris adds. “The approach and method have been appealing to people because they have a broad range of uses and applications.”

The research was supported, in part, by the U.S. Department of Energy; Stanford University; and the National Research Foundation of Korea. 

Beneath the Arctic Ocean is an orchestra featuring natural and human composers, from cracking sea ice and whistling beluga whales to humming shipping-vessel engines. Researchers from MIT Lincoln Laboratory heard some of this cacophony when analyzing data from commercial off-the-shelf sensors that they integrated and deployed in 2024 during the U.S. Navy's Operation Ice Camp (OIC). This past March, during OIC 2026, the researchers returned to the Arctic with a higher-fidelity version of one of the sensors, a geophone, which detects vibrations in the sea ice.

"We're interested in things that make sound underneath the ice," says Ben Evans, a researcher in the laboratory's Advanced Undersea Systems and Technology Group. "For example, our OIC 2024 data contained marine-mammal songs. We need a better understanding of how such signals propagate through ice, and how to distinguish these signals from other sources."

This underwater soundscape is shifting as sheets of Arctic sea ice rapidly break and melt, opening previously impassable maritime routes for military and commercial activity. Determining the unique sound profiles, or acoustic signatures, produced by fracturing ice will enable researchers to develop predictive capabilities for building coastal community resilience, informing geopolitical strategy, and surveilling adversary Arctic activity. The Fiscal Year 2027 Administration R&D Budget Priorities and Cross-Cutting Actions calls for agencies to "prioritize research and associated research infrastructure investments that enhance America's ability to observe, understand, and predict the physical, biological, geologic, and socioeconomic processes and interacting systems of the Arctic to protect and advance American interests and ensure prosperity of America's Arctic residents," and to "invest in R&D that assures America's uncontested navigation and strategic utilization of the Arctic."

Weather woes

Evans and David Whelihan have been trekking to OIC since 2022, advancing their vision to distribute a set of low-cost sensors across the Arctic for continuous monitoring. Hosted by the Navy's Arctic Submarine Laboratory (ASL) every two years, OIC is a three-week event during which U.S. and allied military forces conduct operational readiness exercises. The temporary infrastructure ASL sets up for OIC — a drifting sheet of Arctic sea ice, atop which sits a runway and insulated tents for lodging and command and control — simultaneously enables researchers to test prototype equipment and conduct experiments in an environment otherwise inhospitable to humans. This opportunity is particularly valuable as Arctic monitoring systems are developed in support of U.S. Department of War priorities.

This year proved especially challenging, with back-to-back blizzards creating whiteout conditions. Temperatures persistently plunged to minus 25 degrees Fahrenheit, and winds blew at 25 to 30 miles per hour with 40 mph gusts. Although the team, which also included Ella Wawrzynek and Ryan Saenger, had intended on completing two stints on the ice — one to deploy the sensors and the other to retrieve them after a few weeks — the weather had other plans. Their initial trip to camp was delayed by a week as windblown snow halted all inbound and outbound flights.

While they waited in Prudhoe Bay, Alaska, for the weather to clear, Whelihan was readying another technology they planned to test at OIC: a modem from industry partner Havguard, a Norwegian defense technology startup, that can communicate through ice using magnetic fields (instead of radio-frequency signals, which are rapidly diminished by seawater). However, harsh conditions inside and outside in Prudhoe Bay led to some system failures, and he flew back to the laboratory to fix them. "De-risking and operationalizing critical technology for the warfighter is an important part of what we do," Whelihan says.

After Evans, Wawrzynek, and Saenger arrived at camp on March 7, not a single flight came or left for the next five days. During a "normal" mobilization at OIC, six to nine flights per day are typical. "At times, we couldn't see participating countries' flags on poles roughly 100 feet away from the command tent," Evans says. "We had to put on hats and sometimes goggles just to go between tents, whereas at previous OIC events we walked around with long johns and pants."

The day after their arrival, they loaded their sensors onto a sled, and a field party leader (an expert in Arctic survival) driving a 4x4 vehicle with tracks towed them outside the main camp area. After deploying a quarter of the sensors they had planned, their leader received a call from camp command instructing them to return. The windblown snow was picking up, and they soon wouldn't be able to retrace their tracks back to camp.

A week later, the team had a clear day to retrieve their sensors and fly out of camp. "We laughed, because we could very easily see camp from where we had deployed the sensors," Evans adds.

Magnetic communication

As Whelihan returned to Prudhoe Bay with the fixed communications modem, another blizzard hit. Because the laboratory team had already been to camp once and other research teams needed an opportunity to go onto the ice during the next clear-weather window, the laboratory contracted with UIC Science to conduct the modem experiment. A business unit of the Ukpeaġvik Iñupiat Corp., UIC Science provides logistical and technical support for Arctic research based in Utqiaġvik (Barrow), Alaska, the northernmost U.S. city. Unlike the drifting ice in the open ocean, the ice in Utqiaġvik is primarily landfast, meaning it's fastened, or anchored, to the shoreline or seafloor.

Through UIC Science, Whelihan and Wawrzynek learned how to ride snowmobiles and then drove onto a big lagoon, where they drilled a 2x3-foot hole through 3.6 feet of ice to deploy a remotely operated vehicle (ROV) carrying the Havguard communications modem. The modem is based on a magneto-inductive transmitter (positioned below the ice) and receiver (sitting atop the ice), which are housed within polycarbonate domes to protect the sensitive electronics. The duo had met with Havguard in Norway in fall 2025 to discuss plans for testing the modem in the Arctic, and Havguard in turn built a version to testing specifications. Prior to OIC, Havguard and the laboratory team deployed the modem on a large reservoir in Vermont. While not representative of Arctic sea ice over salt water, this environment allowed them to test their procedures and capabilities.

"Under the Arctic lagoon ice, we placed the ROV, which was also equipped with a Doppler velocity logger, a four-beam sonar system that measures the vehicle's speed and direction," Whelihan says. "We used those measurements as the ROV drove under the ice, plus aerial drone images, to superimpose a picture of the ROV on the site so we could track it and calculate the modem's rate of test-data transfer. We achieved through-ice communication at about 1.2 kilobytes per second on an alpha prototype system that had traveled from Boston to Alaska three times. This result is very encouraging, and the system warrants further development."

In future iterations of this setup, the data could then be relayed out of the Arctic through drones or satellites.

Community connections

While in Utqiaġvik for a week in mid-April, they participated over the weekend in the Piuraaġiaqta annual spring festival, watching a harpoon-throwing contest and proctoring a kids' snowmobile race. And with eighth graders at the local middle school, they discussed their Arctic R&D and engaged them in a game teaching sonar concepts.

"When you embrace this culture of community, you meet lots of interesting people and doors open up," Whelihan says.  

"Connecting with the Arctic community is an important aspect of our work," Evans adds. "Every time we come here, we cross paths with someone we don't expect to, learn about their work, and think about how we may be able to collaborate." For example, at OIC 2024, the laboratory team had met a professor from the University of Maryland at College Park with extensive experience collecting and analyzing cryoseismological data; they now hope to work with him to apply machine learning to discriminate between icequakes and marine mammal vocalizations.

In the lead-up to OIC 2028, the team plans to design, prototype, and test air-droppable versions of some of their sensors while continuing to partner with Havguard on the through-ice communications modem. Their next step is to optimize the modem's packaging to facilitate deployability in the Arctic and integration with the laboratory's sensor suite.

"The through-line in all this work is minimizing boots on the ice," Whelihan says. "Especially this year, we learned that the weather is in charge of our access to the Arctic. We need ways to easily get sensors where we want them and to retrieve the data they collect, even in these extremely challenging conditions."

This work is funded through the laboratory's internally administered R&D portfolio in mission-critical technology (integrated systems area) and the laboratory's Advanced Concept Committee, which funds high-risk, high-reward early-stage research addressing critical gaps in national security technology.

Massachusetts Governor Maura Healey and MIT President Sally Kornbluth will kick off MIT Future Fest, a new annual festival exploring the future of science, technology, art, and design, with “The Future Begins Here” panel, a celebration of Massachusetts innovation, talent and the bold questions shaping what comes next. The event will take place on Wednesday, September 30 at 3:30 PM at MIT’s Kresge Auditorium.

Curated and produced by the MIT Museum, the inaugural MIT Future Fest will take place across MIT’s campus from September 30–October 4, 2026. Governor Healy and President Kornbluth will be joined on the opening panel by Moderna co-founder and Flagship Pioneering founder and CEO Noubar Afeyan, MIT professor and entrepreneur Sangeeta Bhatia, and Bob Mumgaard, CEO and Co-Founder of Commonwealth Fusion Systems. Economic Development Secretary Eric Paley will moderate the discussion, which will explore how public, private, and educational institutions can work together to sustain talent pipelines, turn discovery into impact, and build the future

“Massachusetts is where the future is being invented, and MIT Future Fest is a chance to showcase our leadership in technology and design to the world,” said Governor Maura Healey. “From AI and robotics to clean energy and life sciences, the breakthroughs happening here are changing how we live and work. As Governor, I want Massachusetts to be the place where the best minds from around the world come to study, conduct research, start companies and scale their ideas. Our administration is investing in the talent, research and partnerships that make that possible, and we’re proud to launch MIT Future Fest with MIT.”

“The breakthrough discoveries that shape modern life came from decades of scientists and creators asking fundamental questions about how the world works. MIT Future Fest celebrates that same spirit of curiosity on mission," said MIT President Sally Kornbluth. "When you bring together engineers, inventors, artists, scientists, designers and entrepreneurs, you create the conditions for truly transformative innovation. This festival is an invitation to join that conversation, with the conviction that the future isn't something that happens to us—it's something we create together."

“The Future Beings Here” is one of more than 70 public talks, tours, performances, exhibitions, installations, and open laboratories included in the five-day program. Additional festival highlights and the full programming line-up are available at mitfuturefest.org.