From the basement of MIT’s Building 35 to Monterrey, Mexico, and now beyond. That is the journey of FrED, a low-cost desktop fiber (Fr) extrusion (E) device (D), designed and assembled by students in an educational factory at MIT.

That factory is transforming how manufacturing is taught — replacing textbook learning with hands-on experience in a space where tinkering is encouraged and information flows continuously. Through a collaboration between MIT and Tecnológico de Monterrey (Tec) managed by MIT.nano, FrED has been refined across dozens of graduate theses and undergraduate research stays. It is used to study manufacturing systems in academic and professional courses, and at FrED factories, first established at MIT and now at Tec’s campuses in Monterrey and Mexico City.

“What does it mean to bring the factory to the learner?” asked Brian W. Anthony, MIT.nano associate director and principal research scientist in the MIT Department of Mechanical Engineering (MechE) at the second annual FrED summit in Mexico City. “We have FrED as a process that manufactures a fiber, and we also have the FrED factory that’s an education and practice factory where we are manufacturing a real product. It’s not just a learning factory where we tear apart the product when we’re done. We really ship FrEDs to our online learners, to educators at MIT and Tec, and soon, to new partners around the world.”

Designed from the start for multi-node community scaling, FrED and the FrED factory have created a thriving, collaborative ecosystem for current and future manufacturing engineers. The next step is to expand that ecosystem globally. Announced at the FrED summit by Tec professor Pedro Ponce Cruz, a new FrED factory at Tec’s Saltillo campus will be opening in the next academic year. After that, the team plans to expand to other campuses across the United States and Mexico.

“Together, we are helping build a global engineering talent pipeline,” says Adriana Vargas Martinez, executive director of research strategy at Tec. “Through the FrED and FrED factory initiative, nearly 500 students have already been trained in advanced manufacturing automation, moving from Tec classrooms into research laboratories and collaborative projects with MIT.” 

Discussing FrED and FrED factory’s research impact, she notes 25 publications and seven papers in development. “International mobility has also been an important dimension of this partnership,” she says.

A shift toward modern manufacturing deep-tech themes

FrED’s expansion comes at a time when manufacturing at MIT and across industry is shifting toward smart manufacturing, or Industry 4.0, integrating automation, machine learning, and artificial intelligence. One of MIT’s strategic priorities, the MIT Initiative for New Manufacturing (INM), is working to support new manufacturing research, development of new courses and workforce training, and building of shared facilities to pilot production lines and immersive manufacturing experiences. FrED and the FrED factory are already designed to support these efforts, and at an international scale.

“FrED and the FrED factory is really, I think, solving at least one problem: how we give real, physically meaningful physical context and production-level data, production-level problems in an academic environment that is directly transferable to the knowledge that you need on the factory floor,” says Anthony. It’s difficult to get data out of a real factory, he adds; what FrED offers is physical context crossed with data science, providing an open platform and open data for learning and experimenting.

FrED naturally generates the multi-modal data required for digital twins, analytics, and AI-driven process improvement, turning abstract AI/manufacturing integration into hands-on practice. The next set of research objectives in the FrED factory will focus on developing a realistic and interactive digital twin of the factory, immersive technology for collaborative learning, integrating agentic controllers. They will include new downstream manufacturing processes and machines that take as input the fiber from FrED — all to enhance smart manufacturing education.

These goals will be worked on by MIT and Tecnológico de Monterrey students as part of a FrED factory research stay. This program brings Tec undergraduates to MIT to work side-by-side with MIT students — not observing, but fully integrated into the research team. The students then take what they’ve learned back to Mexico, to enhance FrED factories at their home institution. 

“Beyond the technical side, FrED gave me memories, friendships, and a lot more confidence in myself than I knew I had,” says Naomi Najera, a Tec undergraduate student who completed a research stay at MIT in 2025. “It also gave me a space where I could make mistakes and learn from them. And also to realize how much I can achieve with my team. That human side of this project really changed my whole experience.”

A recent result from this exchange, announced June 23 by the American Society for Engineering Education (ASEE), a paper entitled “Hands-On Predictive Maintenance Kit for Manufacturing Education: An Accessible Experiential Learning Approach,” written by Tec and MIT students, received the 2026 ASEE Manufacturing Division Best Paper Award.

Shifting classroom learning to factory operations

At MIT’s campus in Cambridge, Massachusetts, passersby can look down into the Building 35 basement windows to see a constant flow of activity, materials, and knowledge in the MIT FrED factory. In Mexico, seven cohorts of students over four years each designed a custom version of FrED and built and operated an automated FrED factory production line. Indeed, FrED has restructured how Tec teaches mechatronics and manufacturing systems. “This collaboration integrates research directly into education,” says Vargas Martinez, “combining learning factories and our manufacturing environments with student-centered research.”

The Tec students’ enthusiasm has led to the launch of an Undergraduate Research Opportunities Program-like curriculum (FRAME: Factory-based Research for All in Mechatronics Education) in Mexico, where first-year undergraduates are working alongside graduate level students in the FrED factory. 

“Joining FrED as a first-semester university student has been an amazing opportunity for me to get hands-on experience in real-world projects in areas such as coding, manufacturing, and robotics,” says Katherine Lucia McLean. “It’s helped me grow a lot as an engineering student.”

The FrED factory model forces real leadership behaviors: coordinating multi-station systems, managing bottlenecks, building maintenance logic into the student experience, enforcing quality measurement, and iterating system design year after year. As each class graduates and a new one begins, knowledge is transferred, some of it lost, most of it built upon. In this way, FrED never becomes outdated, as each cohort is reimagining manufacturing technologies and systems for a smarter, more productive factory.

FrED and the FrED factory have momentum. Anthony taught the global capstone course at the Monterrey campus last year, and will expand to teach at all five international Tec campuses in 2027. The FrED Factory Conference will take place at MIT in 2027.

Hydrogels are squishy, bio-friendly materials that are made mostly of water and a bit of polymer. The Jell-O-like substance is available in the form of medical patches, sprays, and glues, and can be stuck to the skin or implanted in the body to dress wounds, affix implants, and encapsulate and release medicine over time. 

For all their sticky, stretchy, and protective properties, hydrogels lack one key trait: breathability. If worn for too long, a bandage or patch can trap moisture and sweat, which can irritate tissues and reduce the effectiveness of any device that a hydrogel adheres.  

Now MIT engineers have come up with a recipe for a hydrogel that is both hydrated and aerated, or permeable to air. The new material is just as soft, stretchy, and robust as conventional hydrogels, but a network of tiny tunnels running through the gel allows air to pass through. 

The aerated hydrogel can be worn for longer periods of time compared to conventional hydrogels, without causing skin irritation. It can also reduce sweat buildup, even during exercise. In experiments, volunteers wore wireless heart monitors that were attached to their chest with the new breathable hydrogel. After working out regularly for 10 days, the volunteers showed no signs of skin irritation, and the heart monitors maintained clear readings.  

The results, which are reported today in the journal Nature, may enable longer-lasting hydrogel products, such as breathable bandages and dressings, cosmetic face masks, and contact lenses, along with better-performing health monitors and implants. 

“Water and oxygen are both essential for life,” says Xuanhe Zhao, the Uncas (1923) and Helen Whitaker Professor of Mechanical Engineering, and a professor of civil and environmental engineering, and medical engineering and science. “Now that we’ve added air to hydrogels, people can find broad applications.”

Zhao’s MIT co-authors on the study include Xiao-Yun Yan, Shucong Li, Won Jun Song, Runze Li, Bastien Aymon, Jingjing Wu, Gengxi Lu, Jiayi Liu, Shu Wang, Eric Lu, Hyunhee Lee, James Zhang, Casey O’Brien, and Zachary Smith, along with collaborators from multiple other institutions.

Breathing through Jello

Water makes up about 90 percent of a typical hydrogel. The rest of the material consists of polymers. When mixed with water in a chemical process known as “cross-linking,” the polymers settle into a sort of scaffold that holds the water in place, forming a gel that’s both squishy and stretchy. But because hydrogel’s composition is mainly water, it’s inherently challenging for any air to make its way through the material effectively. 

“In general, water is not breathable,” co-lead author Xiao-Yun Yan says. “Hydrogel is 80 to 90 percent water, similar to Jell-O. And you cannot breathe through Jell-O.”

Other groups have tried to design air-permeable hydrogels, mainly taking one of two approaches. The first has been to essentially puncture microscopic holes throughout the gel. Such designs are breathable, but only in air. When they are placed in liquid, the holes quickly clog up. 

Researchers have also tried mixing hydrogel with certain polymers, such as silicone, that naturally allow air through. But this approach requires adding a large amount of polymers to the hydrogel in order to create enough permeable space for air to move through the entire gel. These hydrogels end up having a greater balance of polymer to water, making them less hydrated in general. 

Zhao, who has been a leader in the development and application of hydrogels, looked to make a hydrogel that lets air through without losing its water-heavy makeup. 

“We want to have lots of tiny channels to let air through, while also maintaining lots of water in the gel,” Zhao says. “This was a significant challenge, and something that people thought was impossible to do.”

Highways for air

After several years of investigation, the team hit on an ideal recipe for a breathable hydrogel that minimizes the non-water ingredients needed to let air through. In their new study, they report that the key to the recipe is “phase separation.” A common example of this process is the interaction between oil and water. The difference in the two liquids’ phases cause them to instantly separate. When the two are mixed, oil and water glom to their own kind, while avoiding the other. 

Zhao and his colleagues took advantage of viscoelastic phase separation in concocting a breathable hydrogel. For their new design, they mixed their conventional hydrogel recipe with a very small amount of silica aerogel particles, which are essentially “solid-form” air bubbles. 

“They are like boba beads,” Yan offers. “The particles are made of silica, which is hydrophobic, meaning that water does not want to leak through them, so they are very stable in water.” 

And as it turns out, the particles are similar to oil when mixed with water. The researchers found that when they mixed just a small amount of the particles with a solution of the water-heavy hydrogel, the water molecules glommed together, essentially finding each other faster than the less abundant silica particles. This effect of viscoelastic phase separation created large pockets of water and squeezed the silica particles into skinny, interconnected tunnels. The team observed that after a few hours, this effect formed a network of thin and sturdy, silica-skinned tunnels through which air could flow.

“It’s as if the particles formed a network of connected tunnels, like air-permeable highways within the hydrated hydrogel,” says co-lead author Shucong Li. 

Once they confirmed that the network had formed, the team cross-linked the mixture, essentially freezing the gel, and its breathable network, in place. They then tested the gel’s breathability and mechanical performance over multiple experiments, including one in which they asked several volunteers to wear the gel, attached to a wireless electrocardiogram (ECG) monitor, while exercising for 20 minutes. The volunteers also wore monitors with conventional, commercial hydrogel adhesives.

Throughout the workouts, the researchers observed that the breathable hydrogel maintained a strong ECG signal, in contrast to the conventional gel which exhibited significant signal fluctuations.The researchers observed similar results in an experiment with several volunteers who wore the breathable hydrogel and ECG monitor over 10 days. 

“We reliably saw that after 10 days, the quality of the ECG signal is still pretty good, and after you take off the monitor, there were no noticeable blisters or redness on the skin,” Li says. “This indicates healthy skin conditions.”

The team also exercised the gel itself, putting it through 10,000 cycles of stretching and compression. After these tests, they found the gel still retained the network of air channels, maintaining its breathability. 

“After 10,000 cycles, there was less than a 5 percent drop in oxygen permeability,” Li says. “That matters, because even with your heartbeat, your chest continuously undergoes small strains. So we have to make sure this gel is durable for such daily activity.”

Zhao says the new study provides a novel approach for others to fabricate breathable and multifunctional hydrogels, using the concept of visoelastic phase separation as a guide. 

“We’ve discovered that this process can create these air-permeable hydrogels, and we demonstrate one application,” he says. “But we think there can be very broad applications. This is a technology platform.”

This work was carried out in part through the use of MIT.nano’s facilities. This work was supported in part by the MIT Hatsopoulos Faculty Fellowship, the Uncas and Helen Whitaker Professorship, a HEALS seed grant, the National Institutes of Health, the National Science Foundation, and the Department of Defense Congressionally Directed Medical Research Programs.

In 2024, a U.S. government official warned that Russia could be developing a new satellite designed to carry nuclear weapons into space. The statement followed the launch of a suspicious Russian satellite into low-Earth orbit in 2022, just a few weeks before the country’s full-scale invasion of Ukraine.

A nuclear detonation in low-Earth orbit — the region about 100 miles to 1,200 miles above Earth’s surface — would release trillions of highly energetic electrons that would destroy many of the satellites in space, disrupting telecommunications networks, GPS, space-based internet, and more.

The 1967 Outer Space Treaty bans the placement of nuclear weapons in space, but there’s currently no way to verify satellites don’t contain nuclear weapons. In fact, no verification methods have even been proposed in unclassified, peer-reviewed literature.

Now, MIT Professor Areg Danagoulian is proposing a way to determine if a satellite orbiting Earth contains a nuclear weapon. In a new paper published in Nature, Danagoulian describes his idea for a satellite-based sensor system that could orbit close by a suspect satellite and detect neutrons generated by high-energy protons colliding with radioactive material.

In the paper, Danagoulian calculates that a sensor system the size of a large encyclopedia could detect a nuclear weapon with 99 percent accuracy if it orbited within 4,000 meters of the suspect satellite for about a week. He also estimates that the detection time could be cut to a matter of hours if multiple satellite sensors were used or the sensor satellite was able to get within 1,000 meters of the suspect satellite.

“If we eventually have some verification mechanisms for the Outer Space Treaty, that will put pressure on countries to respect the treaty or disclose what they are doing, because they know if they try to violate it, we will find out,” Danagoulian says. “I very much hope this will turn into a real system, or proof-of-concept system, but the goal right now is to get national labs to use this work for their own research, and to get policymakers to seriously consider this technology as a potential part of national technical means.”

Protecting space

In 1962, the U.S. detonated a 1.4-megaton thermonuclear warhead in space, which unintentionally destroyed many of the early satellites of the era. The blast released enormous volumes of highly energized electrons, and many became trapped in Earth’s magnetic field, where they damage any electronics in their path.

“When you have a nuclear detonation in outer space, basically the whole body of the bomb becomes ionized, and nearly every single electron in the weapon’s mass becomes free,” Danagoulian explains. “It gets injected into what’s called the inner Van Allen radiation belt. Once there, the electrons start hitting everything flying through those belts, causing ionization, radiation damage, and more. As you go further out into space, you create these thick belts around Earth populated by highly energetic protons and electrons.”

The 1967 Outer Space Treaty declared space the “province of all mankind” and banned nuclear weapons in space, among other safeguards. It has since been signed by 118 countries including the U.S., China, and Russia.

Monitoring compliance with the treaty has taken on increased urgency since Russia’s 2022 launch of a suspicious satellite, Cosmos2553, which Russia claims is used for surveillance and sensing. However, U.S. authorities believe it may carry components of a nuclear device undergoing testing, with the possible future goal of fielding an actual nuclear anti-satellite weapon. The detonation of a nuclear weapon at that orbit could destroy many of the U.S. reconnaissance satellites, international communication satellite platforms, as well as the Starlink satellites.

“The Russians launched this satellite in a very strange and unusual orbit because it goes through the most hostile environment possible around the planet,” Danagoulian explains. “No one puts satellites there because it’s highly radioactive. Why would you put a satellite in that orbit? Well, that location is likely the best point for trapping electrons if you were to detonate a thermonuclear weapon.”

Danagoulian notes most research on nuclear detection is highly classified, making it hard to know how much progress has been made in national labs. But he wanted to show that scientifically proving the presence of a nuclear weapon in space is possible.

Particle bombardment

The approach Danagoulian developed centers on a reaction known as spallation, caused by highly energetic protons in radioactive environments.

“When an energetic proton slams into elements with a high atomic number, like uranium and plutonium, each proton may knock out something like 40 neutrons,” he explains. “That’s a ridiculously large number. We’re talking about millions of protons per second per square centimeter, with many of them generating 40 neutrons. The question is can you detect some of those neutrons?”

Normal satellites wouldn’t emit nearly as many neutrons, but there are still naturally occurring protons, neutrons, and electrons in the atmosphere, especially in low-Earth orbit. Danagoulian’s concept uses two panels made up of pixels of neutron sensors known as scintillators that interact with radiation and emit light. The panels are sandwiched between synthetic crystal diamond detectors that allow the system to distinguish between neutrons coming from radioactive materials and natural protons and electrons. The two-panel construction then can be used to estimate the direction of the neutron, allowing it to differentiate between natural atmospheric neutrons and those coming from a suspected satellite. 

“Most neutron detectors are very sensitive to protons, so you have to come up with some smart ways to reject protons but keep neutrons,” Danagoulian says. “You also have to tell the difference between naturally occurring neutrons and neutron spallation from the satellite.”

He believes the system, placed inside of an inspector satellite, would be strong enough to survive the harsh environment of low-Earth orbit while also being fast enough to process the protons, electrons, and neutrons that bombard it.

Danagoulian’s calculations on how long the detector satellite would have to be near the suspect satellite give him confidence in the feasibility of the system. If a detector satellite were able to get within 1,000 meters of the suspect satellite, it could accurately detect nuclear weapons in about one hour. That would amount to a single flyby.

Danagoulian calls the paper a feasibility study of the concept.

“I say in the paper this isn’t a completely proven system,” he says. “The purpose of the paper is to show the scientific community that it’s scientifically possible to do this. But there are many more practical considerations to be made to actually build these detectors.”

Danagoulian hopes the study will stimulate further research and development. He is also working with researchers in MIT’s Center for Nuclear Security and Policy (CNSP) to understand the policy landscape around this issue.

If a version of his system is eventually developed, Danagoulian believes it could encourage the nonproliferation that has helped preserve satellites so far. He notes that while adversarial countries are naturally suspicious of each other’s claims, scientific evidence would strengthen trust.

“You can fake intelligence,” he says, “but you can’t fake physics.”

The work was supported, in part, by the National Nuclear Security Administration, the Carnegie Foundation, and Longview Philanthropy.

Last week, national security agencies from the Five Eyes—that’s the rich, English-language-speaking countries club—jointly released a statement warning of the increasing cyber risks of AI models: in particular, their ability to autonomously hack into systems and networks. The statement was more measured than some of the breathless headlines about it, and the advice they gave is pretty much the standard advice everyone gives—albeit with newfound urgency.

Internet risks are nothing new, and cyberattacks—both large and small—have been a significant issue since long before the current crop of generative AI models...

The move deepens what Democrats say is the president's politicization of federal emergency management decisions.

The American Clean Power Association is assessing the policy and tax landscape for renewable energy sources after Republican rollbacks.

If his deregulatory actions survive judicial scrutiny, Aaron Szabo will have played a key role in reshaping EPA’s regulatory power.

The EU’s plan to support livestock farmers recasts a climate problem as a strategic asset.

They say the European Commission should “strengthen the EU ETS and make it future-proof to ensure long-term investment predictability and regulatory stability."

Heat-related damage tends to be less straightforward to measure than losses caused by floods and wildfires, the European Banking Authority said.

Mounting death tolls in the U.S., Australia, Chile and other countries over the past decade have revealed how unprepared many areas are for such infernos.

The figures are the latest evidence of the deadly toll of recent extreme temperatures across Europe.

The Regional Greenhouse Gas Initiative is delivering a windfall that states can use to cut planet-warming emissions — or lower bills.

The state's largest electric utility got approval to adjust power prices in line with a first-in-the-nation law.

Nature Climate Change, Published online: 08 July 2026; doi:10.1038/s41558-026-02690-1

Carbon markets reward the climate mitigation benefits of nature-based solutions but rarely value their broader societal impacts. Creating markets for verified co-benefits, such as coastal protection and biodiversity, could redirect investment towards projects that deliver larger gains for both people and climate.

If you are crossing an unfamiliar room in the dark, you may grope around a bit to get a sense of your space.

But for many animals, feeling out a space comes more naturally. A mouse, for instance, can efficiently navigate in the dark just by grazing its whiskers against walls and other obstacles.

Fan Wang, a professor of brain and cognitive sciences and an investigator at the McGovern Institute for Brain Research at MIT, has discovered how neurons in a mouse’s brainstem use signals from the animal’s touch-sensitive whiskers to estimate an object’s distance from the face.

Her team’s findings, published June 25 in the journal Neuron, unlock key circuitry the brain uses to represent the space immediately surrounding the body.

Mapping space

The circuit the team discovered is part of the brain’s system for creating an egocentric map of space — that is, understanding where things are relative to one’s own body. Neuroscientists know that the brain calls on specialized circuits to understand space in this way, which are different from its system for mapping space using external landmarks.

In their study, Wang and her team explored how the brain maps the space closest to the body, known as the peripersonal space. This is the space in which we move, and it is vital that we understand where things are in relationship to our bodies so we can reach, step, avoid hazards, and otherwise interact effectively with our environment.

Wang says mice were an appealing model for investigating how the brain understands objects’ distance within the peripersonal space, because a rodent’s whiskers seem so much like a built-in set of rulers. These whiskers, which vary in length, are swept back and forth as the animals explore their environment. As whiskers bend and vibrate, the mechanical sensations are relayed to the brain by sensory neurons at their base. Those neurons fire more when a whisker bends close to the face than they do in response to contact near the whisker’s tip, communicating information about the proximity of the touch.

Wang’s team wanted to know if the brain uses these signals to build an internal ruler-like representation of distance more precise than “near” or “far.” To find out, graduate student Wenxi Xiao and Research Scientist Kyle Severson monitored neural activity in a small sensory-processing region in the brainstem where tactile signals from the whiskers first arrive in the brain. They studied what happened there as mice walked on a treadmill while brushing their whiskers against a wall that passed by at different distances.

Many neurons in the region were sensitive to the whisker bending triggered by the wall. Some behaved similarly to the sensory neurons they were getting their information from, firing more when the wall was closer to the face and thus serving as a proximity-based distance code. But other cells were tuned in to discrete distances, firing only when the distance of the wall the whiskers had touched was within a specific range.

The whiskers rule

For some neurons, activity peaked when the wall was 23 millimeters away from the face, near the tips of the longest whiskers. Others responded most when the wall was at intermediate distances. “Each of these neurons represents a specific distance, and together they span the full range reached by the longest whisker, like tick marks on the ruler,” Wang explains. “We call that the map code.”

The team wanted to know how the brain converts proximity signals from different whiskers into accurate map code of object’s distances from the head. “You cannot just listen to individual whisker neurons, because a contact at the tip of a short whisker would be in the middle of a long whisker. You need a brain circuit to build a unified distance map,” Wang says.

Through computational modeling and by exploring what happened when they manipulated neural signaling in specific ways, Wang’s team showed how distances can be calculated by comparing inputs from different sensory neurons. Their findings suggest that each brainstem neuron that makes up the map code receives both direct excitatory inputs from proximity-sensitive whisker neurons and inhibitory inputs from neurons driven by proximity-dependent whisker touch signals.

“Essentially, the inhibitory pathway allows the brainstem to compare two inputs by subtraction,” Wang explains. “If one input signals ‘this is how far it is’ and the other signals ‘this is how far I estimate it to be,’ subtracting one from the other yields an intermediate value. We think it’s a simple and elegant way to transform tactile input into a representation of discrete distance.”

Wang notes that despite their importance, the brain’s body-centered representations of space have so far received little attention from neuroscientists, who know much more about how we understand locations in space relative to landmarks (an allocentric map). She is eager to investigate how the egocentric map code her team discovered is integrated with other brain systems to guide movement, social interactions, and other behavior, and hopes the findings will further exploration from other groups.

The study was funded by grants from the National Institutes of Health.

In today's world, artificial intelligence chatbots such as ChatGPT and Claude can perform many functions, such as composing work emails and planning travel itineraries. These chatbots are systems built around large vision-language models (VLMs): AI trained on a massive dataset that includes books, websites, code, and images. 

The AI algorithms are then refined on massive amounts of human-generated feedback to follow instructions and avoid harmful or unwanted output, and use that "knowledge" to produce text or images based on input from a user. Although chatbots have clear limitations, they can be very helpful for a wide range of tasks, including in some areas that traditionally require specialized skills, like computer programming.

As part of a project for the U.S. Department of the Air Force–MIT AI Accelerator's Phantom Program, U.S. Air Force cadet Joshua Lynch — with the help of his mentor, Laura Niss, a technical staff member in the Embedded and AI Systems Group at MIT Lincoln Laboratory — wanted to determine if, as a complete novice to coding, he could develop a fully functional program. He used a process called "vibe-coding," in which a user relies entirely on prompts to guide a generative AI chatbot to write and refine code. 

His motivation was to empower anyone familiar with the military problem space, regardless of their technical background, to advance their ideas for useful software applications, essentially bypassing the time and cost constraints of the traditional military software development pipeline. Lynch aimed to build his own application while Niss monitored his experience with the technology.

"The Phantom student wanted to see if he could create a useful application through self-identified vibe-coding, without any previous experience," Niss says. "Within this project, I wanted to understand how his perception of AI changed over time with use. We both wanted to understand better where and how AI could be used by nontechnical users in the military."

Lynch set out to see if, starting with no coding skills and using chatbots, he could create an application specific to his type of tactical team to help reduce collateral damage while enhancing survivability in the broader mission. This application would offer capabilities including AI-assisted target recognition; modular intelligence, surveillance, and reconnaissance; autonomous striking; and communication management on the battlefield. 

During the project, Lynch completed several professional development courses in AI and familiarized himself with both military and nonmilitary uses of the technology. For the basis for his code generation, he used the paid models of three AI chatbots: Anthropic's Claude, OpenAI's ChatGPT, and Google's Gemini. Most of this work was done only through the chatbots' main chat function on a web browser, not as an integrated system within a development environment, as is standard now. The final application was produced using Google AI Studio App, which can create applications that interface with the Gemini application programming interface and has AI integrated in the development environment. 

Over three months, Lynch worked with these models to build his application, called the Remote Operating Modular Augmentation Device (ROMAD-AI). During this time, he learned several methods to improve the code output. For example, he often encountered difficulties with the AI chatbots lacking hierarchical focus and modifying unrelated code sections. He discovered it was important to break problems into small parts, frame questions clearly, and steer conversations back on topic when they stray too far from the objective. 

Learning to recognize the chatbots' limitations and effectively work around them took up most of the project timeline. As Lynch gained more experience with the chatbots, limitations in the AI capabilities and time for development caused him to re-scope the project, moving it from an application that could assist on the battlefield to one that could perform basic document processing, such as analyzing tactical maps of battlefields and generating mission-planning documents through an interface with a VLM-powered chatbot. While the resulting prototype did not perform all capabilities Lynch originally set out to include (and in its current iteration was not secure for the desired use case), it proved the capability and usefulness of such an application for service members.

"I was quite impressed with this final product, and it showed me how powerful these systems can be at prototyping designs from nonexperts," Niss says. "I'm now of the opinion that these can be powerful tools for nontechnical experts to convey problems and possible solutions to technical experts, and aid in communicating desired outcomes."

Niss observed the change in Lynch's perspective of AI language models during his experience. After starting with an impressive goal, Lynch gained understanding of the capabilities of current technology and significantly scoped down his expectations by the end of the project period. Measures of his perceptions of the different AI systems over time and across system updates were particularly interesting to Lynch and Niss, with Claude showing more stability than ChatGPT across traits such as likeability, anthropomorphism, and perceived intelligence. Lynch found AI to be a helpful tutor, but noted its inaccuracies on topics he knew well.

The project showed that AI chatbots can empower nontechnical service members to produce viable software applications for their unique problems, although it works better as a prototyping assistant than as a full production tool when handling sensitive information and for critical applications. Improper vetting of code may lead to security risks, as demonstrated by an instance where Lynch didn't realize that the final application was sending the input documents to a Gemini AI model to analyze, rather than parsing the documents locally on his computer. Although AI can generate significant amounts of functional code, code review remains a bottleneck in this space.

"For me, this project reinforced the expanse between experts in different fields," Niss says. "No matter how good AI gets, I think we'll always need to collaborate to get to the best solutions for the most important problems."

Research was sponsored by the Department of the Air Force Artificial Intelligence Accelerator and was accomplished under Cooperative Agreement Number FA8750-19-2-1000.

This blog post is part 1 of a 2-part series. The second part will set out recommendations for companies and policymakers.

Six years ago—one month into a global pandemic—we argued that the automated moderation processes many platforms were rapidly adopting should be highly transparent, easily appealable, and temporary. We warned that "protocols adopted in times of crisis often persist when the crisis is over."

That warning proved prescient. The use of automation and artificial intelligence (AI) to identify, flag, and moderate content has become the new norm—a permanent feature of how platforms govern speech online. In this two part series, we’re take stock of this new norm, and considering what platforms can and should do to ensure that AI serves online expression rather than stifling it.

A brief history of automated content moderation

From spam filtering and keyword blacklists to the hash-matching technologies used to identify child sexual abuse material and terrorist content, automated technologies have been used in commercial content moderation for many years. While these tools have long posed risks to freedom of expression, their use was, for quite some time, relatively limited in scope.

Then, in 2017, a blog post published by Facebook (now Meta) described the company's "fairly recent" use of artificial intelligence to identify, classify, and remove violent extremist content. At the same time, Facebook emphasized caution, noting that it did not want to suggest there was "any easy technical fix."

Just one year later, Mark Zuckerberg appeared before the U.S. Senate's Commerce and Judiciary Committees and disclosed that "99 percent of the ISIS and Al Qaida content" removed by Facebook was flagged by AI "before any human sees it." He also stated that Facebook was "developing A.I. tools that can identify certain classes of bad activity proactively and flag it for our team at Facebook." At the time, we raised concerns about the ethical implications of using AI in this manner.

Then came 2020. The sudden reduction of the human moderation workforce, combined with a dramatic increase in social media use—and with it, a surge in misinformation—created the perfect conditions for platforms to expand their reliance on AI-driven moderation. It quickly became apparent that companies'—and particularly Meta's—approach to moderation during the pandemic represented a backslide in transparency, freedom of expression, and access to remedy. The increased reliance on automation was a significant factor.

The costs and benefits of AI content moderation

We knew in 2020 that the use of AI to moderate content would present problems for online freedom of expression. Today, those problems are well-documented. A 2025 joint declaration by special rapporteurs and representatives of the United Nations (UN), Organization for Security and Co-operation in Europe (OSCE), Organization of American States (OAS), and African Commission on Human and Peoples’ Rights (ACHPR) states:

“The use of AI content moderation can lead to over-removal, discrimination and censorship. Reliance on inherently biased datasets and opaque training processes can amplify pre-existing inequalities, risking homogenisation of expression, and erasure of linguistic and cultural diversity.”

EFF and many of our allies have documented these impacts. For example, our 2019 paper co-authored with Witness and Syrian Archive examined the impact of extremist content regulations—and their implementation through automation and AI—on human rights documentation. A 2020 report from Human Rights Watch highlighted the consequences of these removals, noting: "There is no way of knowing how much potential evidence of serious crimes is disappearing without anyone's knowledge."

The Center for Democracy and Technology's recent series on content moderation in the Global South demonstrates persistent inequities in content moderation of four “low-resource” languages—so-called because the relative scarcity of training data makes it more difficult to develop equitable and accurate AI models for them. 

Content moderation often disproportionately impacts vulnerable and historically marginalized groups, and AI content moderation is no different. GLAAD recognizes the role AI plays in scaling content moderation but notes that “when moderation systems lack nuance, transparency, and human oversight, they can fail to curb harassment and wrongly suppress legitimate LGBTQ content.”

These failures are not incidental. They are a predictable consequence of deploying automated systems to make complex judgments about language, culture, context, and identity at scale.

All of that said, automated content moderation can offer important benefits. The primary one: helping to spare human content moderators who must review content that varies from whimsical to horrific, often for little pay and with devastating mental health consequences. Outsourcing this work to the bots can offer some relief—though it’s worth noting that the humans hired to train the AI models face a similar dynamic.

In addition, AI models could potentially be trained over time to be more precise, accurate, and dynamic, helping to mitigate over-censorship and disinformation. The jury is still out on whether this potential will be realized; what we do know is that new approaches to the persistent problem of over and under-enforcement are desperately needed.

Automated moderation is no longer an experiment

Getting the balance between real costs and potential benefits depends a lot on the details: how automated systems are designed, trained, implemented, and audited.  

Despite advances in the sophistication and scale of automated moderation systems, many of the transparency, accountability, and due process safeguards advocated by civil society, researchers, and human rights experts have yet to be fully realized. At the same time, automated systems have become increasingly central to how platforms enforce their rules and govern online speech.

The question today is not whether companies will use AI to moderate content, but under what conditions they should do so. And now as ever, the answer is not that the public should just trust that platforms’ deployment of increasingly powerful systems will serve, rather than inhibit online expression. In fact, as automated systems become more sophisticated and more deeply embedded in platform governance, the need for transparency and accountability becomes more urgent. 

In the global race to build and dominate the AI industry, it can sure seem like the interests of ordinary people sit last on the agenda. It's just the opposite for EFF. While companies furiously jam AI tools into their veins and your eyeballs, EFF’s technologists, activists, and attorneys have been meticulously cutting through the hype to ensure AI can serve your privacy and free expression. Technology has leaned into a new era, and this summer you can help EFF fight for the people.

JOIN EFF

Over the next two weeks, we’re encouraging you to support the cause as an EFF member for as little as $10 each month. You can get great member swag every year like our privacy puffy stickers, Claw Back t-shirt, and Privacy Badger Crewneck.

Fight mass surveillance! Pictured: Claw Back member t-shirt and Privacy Badger Crewneck.

AI tools—beyond their marketing fluff—demonstrate both incredible potential and real danger. With the support of members around the world, EFF detangles the possibilities from the anxieties and threats with the care and nuance it deserves. In recent months, EFF:

• Mobilized people against the GUARD Act, which would require problematic age verifications systems for AI companions.
• Joined civil society partners to call out a General Services Administration proposal that would make AI tools less safe and less useful.
• Sued for answers under the Freedom of Information Act to uncover how the government is using AI to evaluate requests for medical care.
• Testified before the U.S. Congress Homeland Security Subcommittee on Cybersecurity and Infrastructure Protection.

The scope of AI, both the good and the bad, multiplies every day. If we want the AI-powered benefits of efficiency, scientific discovery, and greater accessibility to knowledge, then we also need strong protections against surveillance, harms to creativity and innovation online, perpetuating systemic bias, and privacy violations now.

With AI taking over the public consciousness, you can be assured that EFF will never stop advocating for you. Together, we can ensure that technology supports freedom, justice, and innovation for all people.

Join EFF

____________________

EFF is a member-supported U.S. 501(c)(3) organization. We've received top ratings from the nonprofit watchdog Charity Navigator since 2013! Your donation is tax-deductible as allowed by law.

When a star dies, a black hole is born. This has been the textbook origin story for most black holes. At the end of a massive star’s life, its outer layers blast away in a brilliant supernova, and its core collapses into a gravitationally tight and dense region, forming a black hole.

Recent discoveries from gravitational-wave detectors have revealed hundreds of merging black holes across the universe. Many of them have been thought to come directly from exploding stars. But black holes can also come from other, smaller black holes. The products of previous black hole mergers can, in principle, merge again, creating a more massive black hole. This alternative, black-holes-birthing-black-holes pathway is known as “hierarchical merging.”

Now MIT scientists are finding that a good number of merging black holes may have indeed merged before. They carried out a new analysis of recent data from the LIGO, Virgo, and KAGRA observatories, containing 155 pairs of binary black holes, and found about 14 percent of merging black holes in the universe may in fact be second-generation black holes that formed from the previous merging of two smaller black holes. 

The results, which the team reports this week in Physical Review Letters, suggest that repeated hierarchical merging is a significant pathway by which black holes form. 

“We’re finding that, for some of these merging black holes, it’s not their first rodeo,” says the study’s first author, Cailin Plunkett, a graduate student in MIT’s Department of Physics. “Overall in the universe, black holes are merging all the time. The question of how often are they repeatedly merging was pretty uncertain. Now we’re seeing a relatively consistent picture where there’s a decent percentage of black holes that are coming from this repeated pathway.”

The study’s co-authors are Salvatore Vitale, associate professor of physics at MIT; Thomas Callister of Williams College; and Michael Zevin of Adler Planetarium and Northwestern University.

Lopsided pairs

When a massive star collapses and dies, the resulting black hole should have very little spin. In addition to losing a huge amount of mass when it explodes, the star should also lose much of its inherent spin, or angular momentum. The black hole left over should then have little to no spin. 

In contrast, when two black holes merge, the collision should create a new, wildly spinning second-generation black hole. 

“They would be spinning very fast, at about 70 percent their maximum possible spin,” Vitale says. 

Scientists suspect that hierarchical mergers occur in dense stellar environments, where stars are so tightly packed together that multiple neighboring stars could die and collapse to form black holes that are then close enough to merge with each other to form second-generation black holes. 

“You might have a ton of stars whizzing around each other, and if some are massive and explode, they become black holes. The black holes continue to whizz around, and can capture each other and merge,” Plunkett says. “This process can repeat potentially ad infinitum, by virtue of the fact that you have a ton of stars and black holes in this really dense environment.”

One sign of a hierarchical merger is that one black hole in a pair of merging black holes has a much higher spin, and higher mass, than the other. Such a lopsided duo would signal that at least one of the black holes came from the collision of two previous black holes. 

In 2024, scientists detected two such lopsided mergers in signals recorded by the LIGO, Virgo, and KAGRA observatories. The observatories detect incoming gravitational waves — incredibly small wobbles in the fabric of space and time — that are the reverberations from distant cosmic phenomena, such as colliding black holes. 

The observatories detected two gravitational-wave signals, labeled GW241011 and GW241110, each of which likely contain a black hole spinning much faster than its partner. The hierarchical mergers were discovered by analyzing each signal in detail to tease out the specific masses and spins of the black holes involved in each merger.

That work inspired Plunkett and Vitale to do a search of similar hierarchical mergers using all the gravitational-wave signals that the observatories have captured to date. 

A pattern of wobbles

For their new study, the team analyzed the LIGO-Virgo-KAGRA Gravitational Wave Transient Catalog 4.0 (GWTC-4.0), which comprises gravitational-wave detections from the observatories’ fourth observing run. Rather than analyze each gravitational-wave signal one by one, which is what scientists did for GW241011 and GW241110, Plunkett and Vitale searched for a characteristic pattern of hierarchical mergers across the data overall, to see if any matching signals popped out.

The pattern they searched for represents a range of orbital “wobbles.” Just before they merge, two black holes spiral toward each other in a disk-like, orbital plane. When the spins of the pair are perpendicular to the plane, this remains relatively steady. But when one or both spins are not perpendicular to the plane, the disk will wobble. The degree to which the whole plane wobbles, or “precesses,” can tell scientists about the balance of masses and spins between the two spiraling black holes. 

Plunkett and Vitale developed a model for the range of wobbling that should be a sign of a hierarchical merger, specifically between a first-generation and a second-generation black hole. 

The team applied the model to the entire GWTC-4.0 catalog, which comprises gravitational-wave signals from 153 black hole mergers, in addition to the signals from GW241011 and GW241110. Their analysis revealed that a number of mergers fit the pattern for orbital wobbling that was likely caused by the colliding of first- and second-generation black holes. 

Specifically, they found that roughly 14 percent of merging black holes in the universe may have merged before, and that these second-generation black holes had very particular masses: Black holes of around 10 solar masses (10 times the mass of the sun) and 30 solar masses were run-of-the-mill star-born black holes, while second-generation black holes had masses of around 20 solar masses or 40 solar masses and above. 

“One of the reasons why the 40-and-above regime is interesting is, stellar evolution theory predicts you shouldn’t be able to form black holes in that mass range at all from just a supernova,” Plunkett says. “We think supernovae from really massive stars end up being so violent that they leave no black holes at all above roughly 45 solar masses. Yet we have seen black holes that are that massive. And the question is: Where did they come from?”

The team’s new analysis provides support for the idea that black holes can form from the repeated merging of other black holes, and that this alternate origin story could explain some of the curious black holes that we can detect today. 

This work was supported, in part, by the National Science Foundation, and the Brinson Foundation.