For decades, neuroscientists have known that specific regions in the brain’s left hemisphere are responsible for processing language. However, a new study by MIT researchers shows that language processing also occurs in many other parts of the brain.

Using functional magnetic resonance imaging (fMRI) data from more than 700 people, the researchers identified 17 additional regions of the brain that appear to play a role in language. These regions are scattered across the brain, including parts of the cerebellum, hippocampus, and cerebral cortex, and they make up about 5 percent of the total volume of the adult brain — about the size of a large strawberry.

“Even though there are all these distant components, it’s pretty restricted in terms of volume. You don’t need that much of the brain to do language,” says Evelina Fedorenko, an MIT associate professor of brain and cognitive sciences, a member of MIT’s McGovern Institute for Brain Research,and the senior author of the study.

Exactly how these regions contribute to language processing is still to be discovered, although the researchers have made some progress toward determining the functions of the cerebellar regions that they identified.

MIT postdoc Agata Wolna is the lead author of the paper, which appears in the Journal of Neuroscience. Other authors include Aaron Wright, a K. Lisa Yang Post-Baccalaureate Research Scholar at MIT; Colton Casto, a graduate student at Harvard University; Samuel Hutchinson, a graduate student at MIT; and Benjamin Lipkin PhD ’26.

Tracking language

The brain’s language processing centers include Broca’s area, first discovered in the 1800s, plus additional regions in the left frontal and temporal lobes of the brain. Scientists have found that some of the corresponding areas of the right hemisphere also contribute to processing language, especially the social-emotional components of language.

There have also been hints that other parts of the brain might be involved in language processing. Early in her career, Fedorenko’s language studies often showed active brain regions outside of the canonical language centers, but she says she was discouraged from including them in her papers.

“When we initially started looking at language, in the first couple of papers, I tried to be comprehensive and include anything that seemed consistent across participants, and there was a huge amount of resistance,” she says. “People would say things like, ‘Well, we know those are not language areas, so please focus on the language areas.’”

In the new study, she and Wolna wanted to revisit those brain scans and see if they could systematically identify language regions outside of the standard language-processing areas.

To do that, they analyzed data from 772 people who had been scanned in Fedorenko’s lab since 2013. Each of these participants underwent a task known as a language localizer, which is used to determine the location of language processing areas for each subject. 

During the test, participants read or listen to sentences as well as sequences of nonwords. For each person, the researchers measure the difference in strength of response when reading real sentences or nonsense sequences. The brain areas that work harder during the sentence condition are considered to be doing something relevant to language, especially if they respond while both reading and listening to sentences.

“It’s a very simple paradigm that lets you identify this core language system in individual brains,” Wolna says.

When searching for language areas, the researchers usually use a relatively strict statistical threshold. In this study, they relaxed the threshold and also used some targeted searches in subcortical areas, in hopes of finding all areas that may contribute to language processing.“We always see this frontal temporal network, but there’s quite a lot of evidence that there are other regions that are also critical for language processing,” Wolna says. “By using a laxer threshold and zooming in on areas with weak MRI signal, we tried to maximize the chances of finding small and weakly responsive regions outside of this left frontal temporal system.”

A widespread network

For about 490 of the participants, the researchers also had data on how their brain responded during a spatial working memory task — remembering the locations of flashing squares on a grid. This task engages a brain network called the multiple demand system, which does not overlap with the core language areas.

This task allowed the researchers to ask whether any of the newly identified language-sensitive regions specifically respond to language and not more general cognitive processes.

Of the 17 new language sites that were revealed by this study, five are located in the cerebellum, which is mainly involved in coordinating the body’s movement. In a study published earlier this year, researchers led by Casto found that three of those cerebellar regions also became engaged during some nonlinguistic cognitive tasks, which was also seen in the new study.

“Those areas that respond to both language and some other tasks could be really interesting and important because they may be doing something like integrating information from different cortical systems,” Fedorenko says.

They also found language-selective regions in the medial frontal cortex, the bottom surface of the left temporal lobe, the hippocampus, and the amygdala. The researchers now plan to further study how these brain regions might contribute to language processing.

“We can now test some ideas from past work, and also more rigorously characterize these regions across different kinds of language manipulations, and different kinds of nonlinguistic tasks, to try to understand what it is that they’re doing,” Fedorenko says.

The research was funded by the Simons Center for the Social Brain at MIT, the McGovern Institute, MIT’s Department of Brain and Cognitive Sciences, and the MIT Siegel Family Quest for Intelligence.

As a new academic year dawns, the MIT-Kalaniyot program is welcoming its second cohort of scholars to campus, expanding an innovative effort to build new connections between MIT and researchers from Israel. 

In fall 2026, MIT-Kalaniyot has 11 new scholars arriving at MIT to pursue research, collaborating with Institute faculty across a wide variety of disciplines. They consist of seven new Kalaniyot Postdoctoral Fellows and four new Kalaniyot Sabbatical Scholars, who are faculty on leave from institutions in Israel. 

It is another step forward for a program which, less than two years ago was still an idea on a drawing board. The project aims to enhance research and create stronger community ties — not only among those connected to the program, but across the MIT campus.

“The goals of the program are to build academic ties between MIT and Israel, alongside a strong, supportive community,” says Or Hen, an MIT nuclear physicist and a co-founder of MIT-Kalaniyot. “MIT has a mission that revolves around research, education, and entrepreneurship, and MIT-Kalaniyot strengthens MIT, to help meet that mission for the world.”

The scholars will be working on a wide range of topics, including mathematics, materials science, behavioral economics, architecture, modern history, chemistry, quantum computing, and computational methods for examining cellular activity.

“We designed Kalaniyot to strengthen MIT’s research and its community at the same time,” says Ernest Fraenkel, a professor of biological engineering and a co-founder of  MIT-Kalaniyot. “We now have scholars in the program working in each of MIT’s five schools. The academic breadth shows our model is working.” MIT-Kalaniyot will also feature its first teaching fellow at the Institute, hosted by MIT’s History program. 

MIT-Kalaniyot was founded by Hen and Fraenkel as a constructive response to discord over conflict in the Middle East. Hen is the Class of 1956 Associate Professor of Physics and associate director of the Laboratory for Nuclear Science; Fraenkel is the Grover M. Hermann Professor in Health Sciences and Technology.

Fraenkel and Hen credit multiple members of MIT’s community and upper administration for backing the MIT-Kalaniyot idea from the start, making it feasible for the program to launch. 

“When we first shared the idea, we were very encouraged by the response from MIT’s senior leadership,” Fraenkel says. “They understood the value of a faculty-led effort, and their constructive response gave us confidence that our approach could be successful.”

“This would be impossible to do the way we’re doing it without the administration’s support,” Hen says. “The program is faculty-led and institution-backed. That’s what you want.”

Hen adds: “I think MIT today is home to one of the most, if not the most, accepting and welcoming communities for Israelis, and I can stand by that statement very strongly. The way our community grew these past years is remarkable.”

Embedded at MIT

MIT-Kalaniyot, named for a well-known flower that grows in Israel and other parts of the region, welcomed its first cohort of scholars to the MIT campus for the 2025-26 academic year. Hen and Fraenkel also give Tal Cohen, an associate professor in MIT’s Department of Civil and Environmental Engineering, substantial credit for developing the concept. 

Scholars at Israel’s nine state-recognized universities are eligible to seek the MIT-Kalaniyot fellowships, which enable research, collaboration, and training at the Institute. The scholars come from a range of academic and personal backgrounds, including both Arab and Jewish citizens of Israel. 

The program is highly competitive, with many more applicants than positions currently available. Applicants are encouraged to identify in advance MIT faculty they would like to work with; accepted applicants then already have a “faculty host” lined up. Many of the new fellows will be working with researchers in established MIT labs, for instance. 

“When they’re here, they are treated exactly like anybody else in an academic unit at MIT and that’s really important,” Fraenkel says. “They’re embedded in these places.”

The program is also intended to generate the kinds of community connections that help scholars flourish, both professionally and personally. MIT-Kalaniyot features weekly lunches, attended by people from the larger community, where scholars can forge connections and friendship. 

The program also features informal academic talks and discussions, with the talks given by MIT researchers both within and outside of MIT-Kalaniyot. Hen, for one, has already seen the benefits of such events; one paper he has recently co-authored directly stemmed from discussions he had at a program event. 

“The range of MIT faculty who stepped forward as hosts has been one of the most gratifying parts of the program,” Fraenkel says. “It shows that this is not confined to one field or one corner of the Institute. It is becoming part of MIT’s broader academic life.”

Adds Hen: “I think it sends a very strong and important message. We’re able to move forward at MIT and build collaborative partnerships with strong ties.”

An additional facet of the program is the potential impact of MIT-based research in practical, tangible ways. One of the 2025 fellows, a leading physician, focused her MIT work on new methods of breast cancer detection, and now, back in Israel, is working to apply those findings in active medical settings. 

Plans for future growth

Having first taken root at MIT, the MIT-Kalaniyot concept is now spreading to other places. In the last two years, Columbia University, Cornell University, Dartmouth College, Harvard University, the University of Pennsylvania, and the University of Southern California have implemented the concept, with other universities in the process of adopting it as well. 

“This national movement all started by replicating the MIT model,” Hen says. “Each university then innovated in their own way. They start from the MIT approach, and then they adapt to what’s happening on their campus. They learn from us, we learn from them, and together we support a broad academic network.”

The progress at MIT and elsewhere has led Hen and Fraenkel to feel optimistic about the ongoing evolution of MIT-Kalaniyot. 

“We started at a tense time on our campus, not really knowing what the future would hold, and it’s exceeded our hopes,” Fraenkel says. “Now we want Kalaniyot to become a recognized center at MIT, funding seed grants for research that wouldn’t happen any other way.”  

While Fraenkel and Hen do not yet have a firm timetable for those developments, they regard them as being realistic. 

“Now we see Kalaniyot as a program that helps MIT well beyond our community,” Hen says. After all, he observes, simply as a vehicle for research, the program has the potential to provide added capacity for MIT, as well as the further connections to top scholars being generated by the effort. 

Indeed, Hen reflects, he is motivated the question: “How do we best support MIT in realizing its mission for the world?” Overall, he says, “I think that’s the ultimate goal of Kalaniyot. We do it in one way, other people can do it in other ways, and as long as you do net good, and support the MIT mission, we value and treasure that, and just want to be part of it.”

“I really believe this is the DNA of MIT,” Fraenkel says. “We’re all about finding practical solutions to society’s biggest problems. Kalaniyot brings extraordinary people here to do exactly that, and the whole Institute is stronger for it.”

Three teams comprising 35 students across eight different MIT departments and Wellesley College have been at work since fall 2025, designing critical early infrastructure elements that a moon base would require. This June, their designs were recognized with five awards at NASA’s 2026 Revolutionary Aerospace Systems Concepts — Academic Linkage (RASC-AL) Forum. 

Among 75 submissions and 14 finalists, the MIT teams earned first and second place in the competition, as well as three best-in-theme awards. The Exploration-Class Lunar Integrated Power SystEm (ECLIPSE) team won first place overall and first in its theme category, lunar surface power. The communications and navigation constellation team, MELIORA, won second place overall and first in its theme category on Mars communications, position navigation and timing, which included a strategy for proving the design at the moon. And CHEESEBURGER, a campaign to mine and process lunar regolith into oxygen, metals, and bricks, won first in its theme category, lunar technology demonstrations. 

“NASA spent the spring telling the world what critical early infrastructure their upcoming permanent moon base will need,” says George Lordos, a research scientist and lecturer in the Department of Aeronautics and Astronautics (AeroAstro) and in System Design and Management (SDM), who co-advised all three teams. “Over 30 MIT students spent this academic year designing much of the moon base — systems for generating, storing, and distributing power; robust systems for positioning, navigating, and communicating; and early experiments with essential technologies to live sustainably off the moon’s own dirt.”

A power grid for surviving lunar night and winter

The hardest constraint on NASA’s moon base is staying powered, because a failure in life-support power would doom the crew within hours. ECLIPSE is a reference design for a lunar grid engineered to stay up for more than 99.995 percent of the time — fewer than 27 minutes of downtime a year in the worst-case scenario, the standard demanded of the most critical data centers on Earth. It pairs two power sources that fail in different ways: banks of 20-meter solar masts in the sunlit highlands near the south pole, and, for the roughly 18-day stretch each year when the sun drops below the horizon, a pair of buried 20 kilowatt microreactors the team named CARROT, (Compact Autonomous Regolith-shielded Reactor Operating for Ten years). The CARROT reactor, a novel design developed independently by the ECLIPSE team, ended up being similar in design to NASA’s SR-1 reactor for the 2028 mission to Mars, both aiming to maximize speed-to-deployment. 

“Burying each reactor 1.3 meters down shrinks the keep-out zone from kilometers to meters, so crews can work nearby, and it saves tons on required shielding mass,” says Taylor Hampson, a PhD student in the Department of Nuclear Science and Engineering and ECLIPSE team co-lead.

The full design delivers an initial 120 kilowatts using a grid of buried aluminum cables and shielded direct-current power equipment. Laser-equipped rovers provide “Frontier Power” capability, beaming up to 10 kilowatts to sites beyond any cable, from a shadowed crater to a new outpost before its own grid exists. Patrick Riley, a graduate student in the Department of AeroAstro and ECLIPSE team co-lead, says the design’s point is to put reliability ahead of mass: “We sized it so the most likely failures never reach the moon base inhabitants, and so it scales from a first crew of six up to industrial demand without interrupting a commercial lunar economy.”

A network for exploring the moon and Mars, and calling home

MELIORA acts as the base’s relay and GPS. Although RASC-AL framed the communications, positioning, navigation, and timing competition sub-theme around Mars, the team also proposed a plan to validate their design in lunar geometry first, in step with the agency’s strategy to prove technology on the moon before extending it to Mars. To find the best design, the team ran a trade study across 5,764 candidate constellation geometries. The result grows from an initial three satellites to 23, returns more than 100 megabits per second to Earth-orbiting data networks over free-space optical links, and pins a user’s position to within 10 meters. For the Mars design, four relay satellites parked at gravitationally stable Lagrange points keep the link alive even during solar conjunction, the weeks when the sun sits between the two worlds and ordinarily cuts communication. On the surface, a user needs only a portable radio terminal and a chip-scale atomic clock — a timekeeper the size of a matchbox. 

“You should never have to think about whether the network is there — it just is, the way you don’t think about a cell tower,” says Ekaterina Tiukhtikova, an undergraduate studying both AeroAstro and electrical engineering and computer science (EECS), and a MELIORA team co-lead. “We put almost all the complexity up in orbit, so everything on the surface stays portable and simple,” adds Clayton Lieberman, a graduate of the SDM program and team co-lead who wrote his thesis on MELIORA.

Making oxygen, metal, and bricks from lunar dirt

After power and communications, the third essential pillar of a lunar base is living off the land. The moon’s own regolith can supply oxygen to breathe and burn, metal to build with, and shielding to hide behind for protection from deadly radiation. CHEESEBURGER is a campaign of five robotic payloads that prove the supply chain one link at a time, followed by integration of the five into the first end-to-end lunar industry. 

The payloads carry a kitchen’s worth of names: SWISS prospects for the richest ore, BRIOCHES digs and sorts the regolith, BACON casts it into bricks, GRILLED MEAT melts it electrically to pull out metal and oxygen, and AVOCADO is the robotic builder that stacks the products into structures, including interlocking Moon BRICCSS that shield a habitat from radiation. The food theme was born during a January team outing at Sandwich, Massachusetts. “Naming the prospector SWISS and the metal extractor GRILLED MEAT turned a wall of acronyms into something the whole team could enjoy,” says Cesar Meza, a graduate student in AeroAstro and CHEESEBURGER co-lead. “It sounds like a joke until you see that each acronym clearly describes a serious piece of hardware doing one job in the pipeline.”

Thirty students, eight departments, and three teams for one moon base

More than 30 students contributed across the teams, from AeroAstro, SDM, Nuclear Science and Engineering (NSE), EECS, Mechanical Engineering (MechE), the Technology and Policy Program, the MIT Sloan School of Management, and Earth, Atmospheric and Planetary Sciences (EAPS), along with a student from Wellesley College. Several student mentors and faculty advisors worked across more than one team, which is why ECLIPSE’s grid is sized to power CHEESEBURGER’s processing, CHEESEBURGER’s regolith handling is used to bury and shield ECLIPSE’s grid, and all three projects are designed to translate moon base lessons for a future mission to Mars. The teams were advised by Olivier de Weck, the Apollo Program Professor of Astronautics and Engineering Systems and interim department head of AeroAstro, who led ECLIPSE; Kerri Cahoy, the Sheila Evans Widnall Professor of Aerospace Engineering, who led MELIORA; Jeffrey Hoffman, professor of the practice in AeroAstro and a former NASA astronaut, who led CHEESEBURGER; Koroush Shirvan, Atlantic Richfield Career Development Professor in Energy Studies in Nuclear Science and Engineering, who co-advised ECLIPSE; and Lordos, who co-advised all three. Much of the day-to-day mentorship work is led by PhD student volunteers and runs through the MIT Space Resources Workshop, which Lordos founded in 2019.

“The winning teams demonstrated how academic innovation can support Artemis mission goals,” says Daniel Mazanek, RASC-AL program sponsor and senior space systems engineer at NASA’s Langley Research Center, in NASA's announcement of the awards. “Their work highlights the important role student research plays in shaping future space exploration.”

NASA expects astronauts living on the lunar surface for months at a time by the early 2030s — the window ECLIPSE, MELIORA, and CHEESEBURGER were designed for. The picture the three teams had worked toward is unified: a crew at the lunar south pole, the lights on through the winter night, the network always up, and the first oxygen and bricks coming out of the ground beneath them. 

“A permanent base is no longer a slide in a strategy deck; NASA begins landing the first elements in 2027,” says de Weck. “Studies like these three let the agency see, before the concrete sets, how its power, communications, and resource choices depend on one another. That is precisely when independent, integrated architecture work has the most influence on the real plan.”

RASC-AL is administered by the National Institute of Aerospace on behalf of NASA. MIT has a long record in NASA’s student design competitions, with recent winning teams including the  HYDRATION Mars water production system, the Pale Red Dot Mars homesteading architecture, the deployable lunar tower MELLTT, the MARTEMIS lunar Mars analog campaign, the MAPLE autonomous lunar robot pathfinding system, the CERBERUZ lunar recycling project, and the THERMOS cryogenic fluid management system. This work was supported in part by NASA, the Massachusetts Space Grant, MIT AeroAstro, and the MIT Space Resources Workshop. One student was supported by a NASA Space Technology Graduate Research Opportunity Fellowship.

The full teams:

ECLIPSE — Team leads: Taylor Hampson (graduate student, Nuclear Science and Engineering) and Patrick Riley (graduate student, AeroAstro). Reactor team: Liliana Arias, Sydney Menne, Julian Rocher and Pavel Shilenko (graduate students, NSE). Power management and distribution team: Evrard Constant and Mary Foxen (graduate students, AeroAstro), Janhavi Joglekar and Asma Patel (undergraduate students, AeroAstro). Solar and architecture team: Zachary Dawson (graduate student, System Design and Management), Sreeja Akula and Ian Jimenez (undergraduate students, AeroAstro; EAPS), Yohan Lim (graduate student, AeroAstro/Technology and Policy Program), CJ Taglienti (graduate student, AeroAstro/MBA). Student co-advisors: Yana Charoenboonvivat, Lanie McKinney (AeroAstro), Palak Patel (MechE). Industry mentor: Sully Marigliano-Crevecoeur (Technetics). Faculty: Olivier de Weck (lead) and Jeffrey Hoffman (AeroAstro), George Lordos (AeroAstro and SDM), and Koroush Shirvan (NSE).

MELIORA — Team leads: Clayton Lieberman and Katiyayni Balachandran (System Design and Management), Ekaterina Tiukhtikova (undergraduate, AeroAstro and EECS), Celvi Lisy (AeroAstro). Team members: Thomas Harrington and Zachary T. Barnes (SDM), Asael Acosta (undergraduate, AeroAstro). Student co-advisor: Lanie McKinnery (AeroAstro). Faculty: Kerri Cahoy (lead), Jeffrey Hoffman and Olivier de Weck (AeroAstro), and George Lordos (AeroAstro and SDM).

CHEESEBURGER — Team leads: Cesar Meza (graduate student, AeroAstro) and Elizabeth Romero (undergraduate, AeroAstro). Team members: Rachel Dunphy, Shreya Kothnur, Hailey Polson (undergraduates, AeroAstro), Christopher Kwon, Jose Soto, Lanie McKinney (graduate students, AeroAstro), Marvin Martinez (undergraduate, MechE), Ananda Santos Figueiredo (graduate student, Technology and Policy Program), Evangeline Haiqi Wang (undergraduate, Computer Science and Psychology, Wellesley College). Faculty: Jeffrey Hoffman (lead) and Olivier de Weck (AeroAstro), and George Lordos (AeroAstro and SDM).

An MIT-led research program aimed at creating future microsystems capable of sustainably transmitting data with greater bandwidth and higher efficiency than is possible today has made several significant advances since it was established in 2022. 

These include the invention of devices within systems that can much more easily integrate electronics — manipulating data with electricity — with photonics, which does the same with light. The microsystems, the first of their kind, also promise to be cost-effective because, among other advantages, they can be manufactured using existing equipment in traditional electronics foundries and packaging houses.

“Our disruptive electronic-photonic integrated solutions will enable us to leap from [transmitting data at] hundreds of terabits per second to greater than 1 petabit per second,” said Anu Agarwal, who leads MIT’s FUTUR-IC, at an April webinar titled, “Shaping the Future of Semiconductors: Power, Performance, and Possibility.” The event was sponsored by the MIT Industrial Liaison Program and Startup Exchange.

An advanced system using co-packaged optics can provide improved bandwidth and energy savings compared to what is used today, which is electronics-only or pluggable optics.

Toward sustainability

The microchips behind everything from smartphones to medical imaging can be traced to about 500 megatons of carbon dioxide-equivalent lifetime emissions in 2021, and every year the world produces more than 50 million tons of electronic waste. Further, the huge data centers necessary for complex computations like on-demand video are growing, and will require close to 10 percent of the world’s electricity by 2030.

“This is neither scalable nor sustainable, and cannot continue,” Agarwal has reiterated over the years. FUTUR-IC, funded by the National Science Foundation Convergence Accelerator, was created to address these resource-efficiency issues.

For example, integrating photonics with the electronics that underpin today’s microchips could address energy use because the transmission, or communication of data, using light is much more energy efficient. “Our mantra is to use electronics for computation and photonics for communication to bring this energy crisis under control,” says Agarwal.

Currently, however, it is difficult and expensive to connect electronic chips with their photonic counterparts within a single package. That’s partly because the supply-chain ecosystem for co-packaged optics is still immature.

New devices

Enter two new devices developed through FUTUR-IC aimed at making it easier — and less expensive — to integrate photonic chips with microchips. One, the evanescent coupler, was featured on the cover of Advanced Engineering Materials last year. Another, known as the graded index coupler (GRIN), was reported in the March 2026 print issue of the Journal of Physics: Photonics. 

A third new coupler was developed by an MIT team led by Professor Juejun Hu of the Department of Materials Science and Engineering. It was reported in a 2023 issue of Laser & Photonics Reviews. That work was supported by the Department of Energy. 

The three couplers are the first optical equivalents of “solder bumps,” or the tiny dots of metal that allow chip-to-chip or chip-to-substrate connections for electron flow. Until this MIT work, there were no analogous “optical bump” options for photonics.

And if photonics is to be integrated with electronics, “you’ll need both metal bumps and optical bumps, because there are devices on your photonics chip that will require both an electrical signal and an optical signal,” says Drew Weninger PhD ’25, first author of the papers on both the evanescent and GRIN couplers. Weninger is now at the National Institute of Standards and Technology.

As with electronics, many options of optical bumps will be necessary, as “each type has substantial trade-offs,” wrote Weninger and colleagues in a review article in Nature about coupler advances published earlier this year.

For example, the GRIN coupler can be used over a wider spectrum of light than is possible with the evanescent coupler, Weninger says. The evanescent coupler, however, is easier to fabricate and can be packed in tighter to form a higher number of connections.

Additional advances

FUTUR-IC is organized into three dimensions: Technology (the coupler work is a good example), Value Chain Innovation, and Workforce. 

Under the Value Chain sector, researchers developed a new tool to support companies’ decisions toward sustainability. Earthster provides a visual model for quickly determining the energy, materials usage, and environmental sustainability across a company’s products. For example, says Agarwal, “looking at [Earthster], a supplier can tell right away their hot spots for carbon emissions, and start working to minimize them.”

FUTUR-IC has also developed several programs aimed at developing a future workforce for next-generation microchips. For example, “it is introducing an online course on semiconductor resource efficiency,” Agarwal says. “We also offer gamified digital learning and problem-based learning, plus a summer academy and a hands-on bootcamp.” For K-12 awareness, FUTUR-IC has created TED-Ed videos.

Agarwal concluded her April webinar by acknowledging the range of industries FUTUR-IC aims to help. “If you’re a packaging vendor, a materials vendor, or you are in the supply chain for data centers, FUTUR-IC can provide value.”

Additional authors of the paper on the GRIN coupler are Agarwal; Lionel Kimerling, the Thomas Lord Professor in the Department of Materials Science and Engineering; Christian Duessel BS ’25, now at SiLC Technologies, a silicon photonics company; and Samuel Serna, professor of physics, photonics, and optical engineering at Bridgewater State University.

Additional authors of the Nature review paper are Serna; Luigi Ranno PhD ’25, now at Ayar Labs; Kimerling; and Agarwal.

The deployment of automated software systems called AI agents has recently exploded. A November 2025 report by MIT Sloan School of Management and Boston Consulting Group found that 35 percent of surveyed businesses had already deployed AI agents, while another 44 percent planned to implement agentic AI soon. 

To understand the fundamentals and potential impacts of these increasingly popular tools, MIT News spoke with Phillip Isola, an associate professor in the Department of Electrical Engineering and Computer Science (EECS) and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL), who studies the intelligence AI agents possess, as well as the underlying models and mechanisms that power agentic AI systems.

Q: What is agentic AI and how is it different from generative AI models like ChatGPT and Claude?

A: Agentic AI is AI that takes actions in the world. These actions could be a physical action, like robotic manipulation, or a digital action, like booking a flight. On the other hand, we think of generative AI as making up stories, poems, art, and images, rather than taking actions for us. 

The word “agent” is just a brand name. It usually means AI that is going to help people interact with an application, a website, or the physical world. Most agents we encounter today are digital agents, like customer service agents you can talk with about product complaints. 

Most companies that offer agents use the same few AI models under the hood and give them the ability to take actions and remember what happened. An agent starts with a fundamental generative AI system, like Claude, at the core. Then companies put different wrappers around that foundation model for their product or application. Those wrappers might be specific tools that agent can use, and those tools depend on the application. Maybe the agent has access to a calculator so it can solve math problems, or maybe it has access to a more complicated hard drive and operating system so it can remember a firm’s financial data and past business negotiations. 

The biggest challenge in developing agentic AI comes from a lack of training data. If I want to create a system that can go online and book a flight for me, that seems pretty simple. But we don’t have a lot of data that spells out exactly how to do that — where to move the mouse, which buttons to click on, what to do if something goes wrong, or how to call somebody and negotiate about the price of the airline ticket. One way to train a system like this is to have the AI agent visit airline websites, try things out, and see what works and what doesn’t work. These environments are hard to model, so often the agent must learn by trial and error.

Q: What are some promising applications of agentic AI?

A: I think the area where we’ve seen the most success has been with coding agents. This is something that evolved from generative AI. People trained language models on code, and then they can predict what a human would do to solve a coding problem. In addition, an agent can learn to do this by going through a feedback loop where it tries out different solutions and checks to see if it got the answer right. As long as it can check the answer, the AI agent can perform this trial-and-error loop until it figures out a good strategy.

But there is always a balance between automating decision making versus simply assisting and informing humans. Analytical AI methods, like the systems that help predict possible outcomes of decisions, are not agentic in nature, but are very informative to human decision-makers. For cases that are either high-stakes or safety-critical, like medicine, security, high-level business policies, etc., the technology might not be ready for AI to completely automate those processes, or we might not even be comfortable with that.

Q: Are there risks we should be thinking about when using AI agents?

A: One big risk area comes from the fact that it is often very easy to get agents to do certain types of work for you. With coding agents, you can “vibe code” and just ask the agent to make a code for you, so you don’t have to do the hard work yourself. There is a big risk that, because it is so easy, people will not put enough effort into verifying that it is doing the right thing. Bugs will be introduced, private data will get leaked — this is already happening.

Agents aren’t perfect, in the sense that they might make mistakes because they are not well-trained and don’t know what to do. But even if they are very competent, if a human doesn’t use them appropriately or gives them an instruction that is too vague, the AI agent could make a mistake because the human made a mistake. If humans are less involved in thinking through all the consequences, I think we might be more prone to making those mistakes. 

An additional aspect is the risk of de-skilling. It is unclear how far this will go, but when we are relying on agents to do our homework, our coding, and our math, we might lose the ability to do that ourselves, and we might lose that ability too soon because the technology is not yet ready to fully automate those processes.

Q: What does the future hold for agentic AI?

A: What we think of now as agentic AI refers to large language models using tools to interact with digital and physical systems. One obvious limitation is that, under the hood, these have the architecture of a language model and are trained on text data. To make even more powerful AI agents, we might need to model videos, physical forces, time series, radar scans, and other modalities. We might need to have models with fundamentally different architectures that can handle continuous data, high-dimensional data, stochastic data, and so on. 

But, on the other hand, maybe an extremely good coding model could act as a puppeteer to interface with sensors, actuators, and web APIs? Perhaps, once you have a super-smart reasoning system that understands math, language, and code, you can give it a camera and a keyboard and it will figure out what to do in the spatial domain. Is the next wave of AI just going to be Claude with sensors, actuators, and tools, or is it going to be something built in a new way from the ground up? That’s the big question a lot of people in AI are grappling with right now.

The Financial Times has a good article on how AI is changing the capabilities of video surveillance, with information from both Israel/Iran and Russia.

I wrote about this sort of thing a few years ago, how AI enables mass spying in the way that computers and networks enabled mass surveillance. The interesting development in the article is that AI allows people to ask natural language questions about video footage to AIs—and AIs can answer them.

In contrast with older tools restricted to a few dozen preset searches, these new tools allow an almost unlimited range of enquiries by enabling language-based searches on video...

The bank said it would no longer abide by its commitment to direct 45 percent of its financing to climate-related projects.

After forcing a Colorado coal unit to stay open, Energy Secretary Chris Wright says talks are underway about cost allocation.

The settlement agreement provides a partial reimbursement to Duke Energy for its offshore wind lease.

"It came to my attention because I noticed they were arresting people for fixing their car," the president said Monday.

California's biggest oil producer is getting into new lines of business: carbon capture and storage and data centers.

The powerful center-right group’s position will set the tone in Parliament as lawmakers debate the future of the ETS.

Some countries desperately want to gut the Emissions Trading System, others are determined to protect it. Ireland has its work cut out.

Athens is paying subsidies to fishermen to try to contain the proliferation of the marine pest.

But the 2028 deadline leaves manufacturers little time to scale up electric two-wheeler lineups, dealerships and charging networks at once.

This Pride, we’re answering all your digital rights questions in season two of our initiative, LGBT Q&A. 

You Asked: I live in the UK, and we have age verification now on a bunch of websites (including Reddit) and now on iPhones. Can you explain what sort of data companies are actually collecting when they check for age and whether there are any real threats to my safety? 

EFF’s Answer: Age verification is a process where a website or service checks your age to determine whether a user is over a certain age, in the UK this age is 18. 

As of July 2025, all platforms in the UK that host content considered by the UK government and the country’s telecommunications regulator Ofcom to be harmful are legally obligated to check that their users are over the age of 18. If not, users cannot access the content. 

There are various privacy implications for data sharing with age verification. Unfortunately, because services may use different methods to verify users’ ages, you’ll usually have to do a little digging to learn how each provider you have verifies their users, and consider what information might be harmful to your personal safety: 

• The data itself: What info does each method require users to disclose?
• Access: Who can see the data during the course of the verification process? Does anything other than the age result leave your phone or device? Is the provider told your date of birth, or just if you’re over 18? Which third party services see the information you send?
• Retention: Who will hold onto that data after the verification process, and for how long? Sometimes it’s deleted immediately. Sometimes it hangs around forever, waiting for a data breach.
• Audits: How sure are we that the provider’s stated claims around data access and retention will happen in practice? For example, are there external audits confirming that data is not accidentally leaked to another site along the way? Ideally these will be in-depth, security-focused audits by specialized auditors like NCC Group or Trail of Bits, instead of audits that merely certify adherence to standards. 
• Visibility: Who will be aware that you’re attempting to verify your age, and will a third party provider know which platform you’re trying to verify for? Will they hang onto that data to build a profile of you?

Last year, Ofcom outlined a number of methods for online services and platforms to check users' ages. Let's look at some methods in more detail. 

Facial Age Estimation 

First up we have facial age estimation, where you show your face via photo or video, and a technology provided by a company like Yoti or Persona analyses it to estimate your age. Most of these third-party verification services upload your photo to their servers during this process. Yoti claims that “as soon as an age has been estimated, the facial image is immediately and permanently deleted.” 

You might not want to use facial age estimation if you’re worried about a current picture of your face accidentally leaking—for example, if elements in the background of your selfie might reveal your current location. Some services like k-ID and Private ID will analyse your face directly on the device, so only the age result will leave your phone. 

If you do choose (or are forced to) use the face check system, be sure to snap your selfie without anything in the background that you'd be concerned with identifying your location or embarrassing you, in case the image leaks. 

Photo-ID Matching

Photo-ID matching checks whether your photo matches a document that confirms your identity, such as a driving license or passport. This is usually considered the most sensitive, since your ID has quite a bit of information on you. For example, if you upload an image of a document that shows your face and age, and an image of yourself at the same time, these are compared to confirm they match. Like with facial age estimation services, you’ll usually be sent to a third-party provider, such as Yoti or Incode. You’d hope that they’d delete the data immediately, but that’s not always the case. Incode for example doesn’t automatically delete the data you give it once the process is complete; though if you’re reaching them through TikTok, TikTok does claim to “start the process to delete the information you submitted,” which should include telling Incode to delete your data once the process is done. 

If you want to be sure, you can ask Incode to delete that data yourself. But you’re relying on a service you don’t generally have a choice about doing the right thing, and we’ve already seen how that can fail. A previous system that Discord used to verify age had you send a picture to their general help forum, where all of the IDs sat around forever, until they got exposed in a massive data breach. Discord no longer uses that system to verify users’ ages. So, it might be fine, but unless you look into the exact company and all their practices, it’s hard to know. You can check out EFF’s guide for a few of the major platforms. 

Open Banking

Next is open banking, where you give permission for the age-check service to securely access information from your bank about whether you are over 18. The age-check service then confirms this with the online service. The user's full date of birth is not shared. Credit card age checks are also used for pornography services, where you provide your credit card details and a payment processor checks if the card is valid. As you must be over 18 to obtain a credit card in the UK, this shows you are over 18 and can therefore access a service.

Email Verification 

Email-based age estimation is also quite prevalent, where users provide an email address, and a third party technology analyses other online services where it has been used—such as banking or utility providers—to estimate your age. That third party will aggregate some data on you in the process, but the only new information they’ll find out is that you want to verify your age using a particular email address.  

Mobile Operator Checks

Mobile network operator age checks give your permission for an age-check service to confirm whether or not your mobile phone number has age filters applied to it. If there are no restrictions, this confirms you are over 18. 

There is no perfect, privacy protecting verification service

Unfortunately, none of these verification options are perfect in terms of protecting information, especially when this is compounded by the additional risks that LGBTQ+ people face with data sharing. The data can reveal someone’s sexual orientation, gender identity, or HIV status that can be used by employers, governments, family members, scammers, or bad actors to inflict harassment, discrimination, arrest, or violence. 

There is still no widely available way to verify age online without compromising privacy—but even if there were, broad restrictions on social media will inevitably limit access to lawful speech, and valuable online communities, and arts and culture. These are just a few of the reasons that EFF is against age-gating mandates and is working to stop and overturn them in the UK and around the world.

Nature Climate Change, Published online: 30 June 2026; doi:10.1038/s41558-026-02687-w

The ocean absorbs a vast amount of carbon dioxide, mitigating climate change. The projected decrease in this absorption is often attributed to a global-warming-induced slowdown in circulation, but analysis using a mechanistic carbon decomposition and attribution framework reveals that the carbon cycle response depends on the emissions trajectory — with warming dominating under low-emission, high-mitigation scenarios.

Nature Climate Change, Published online: 30 June 2026; doi:10.1038/s41558-026-02674-1

Understanding the socioeconomic impact of climate change at fine scale is essential for promoting real-world actions. Here I look back at a 2018 paper that disaggregated the global economic damages from climate change and discuss how high granularity analyses advance climate impact research and policy progress.

Nature Climate Change, Published online: 30 June 2026; doi:10.1038/s41558-026-02675-0

The scale of carbon dioxide removal (CDR) could determine the extent of co-emitted air pollutants in net-zero scenarios and potential health impacts. By linking a series of models and datasets, researchers find that low-CDR pathways lead to a more equitable distribution of health benefits across the USA.

The Antarctic ozone hole was discovered in 1985, when scientists observed a severe depletion in the Earth’s protective layer of stratospheric ozone. Industrial chemicals known as chlorofluorocarbons (CFCs), then widely used as refrigerants, propellants, foam-blowing agents, and solvents, were at the root of the ozone depletion. After concerted global effort to phase out the use of CFCs, ozone today is recovering, especially in the Antarctic. 

The discovery of the ozone hole was possible thanks, in part, to the measurement tools that were available at the time. Advances in those tools, along with satellites and other monitoring technologies, have since allowed scientists to track ozone’s recovery. 

But what if today’s tech was available much earlier? Would scientists have been able to spot even earlier signs of human-induced ozone depletion? And if so, when would those first signs have popped up, and where? 

MIT scientists now have some answers. The team, led by atmospheric chemist Susan Solomon, has carried out a thought experiment in which they consider a hypothetical world where today’s atmospheric monitoring capabilities were available throughout the last century. In this scenario, they simulated the atmosphere’s chemistry through history and discovered not only when the earliest sign of ozone depletion would have been detectable, but also where, and why. 

In a study appearing today in the Proceedings of the National Academy of Sciences, the scientists suggest that the first signs of ozone depletion appeared as early as 1957 — about 30 years before the ozone hole was discovered. And, this first signal of ozone loss popped up not in the Antarctic, but in the upper stratosphere of the tropics. What’s more, the cause of this early depletion was not due to CFCs, but to another industrial chemical: carbon tetrachloride. 

“What we’ve learned from textbooks is that CFCs result in ozone depletion,” says the study’s first author, Jian Guan, a graduate student in MIT’s Department of Earth, Atmospheric and Planetary Sciences (EAPS). “It turns out there was another compound that caused ozone depletion much earlier than CFCs. This was a big surprise.”

For Solomon, who was an early pioneer in the study of ozone’s effects on the atmosphere, and who was the first to show that CFCs were the main agent eroding Antarctic ozone, the new results were a complete shock. 

“The fact that ozone depletion would have happened as early as the late 1950s, which is much earlier than I would have thought, just absolutely blew my mind,” says Solomon, the Lee and Geraldine Martin Professor of Environmental Studies and Chemistry at MIT. “This study shows it’s really important to keep monitoring so that we can fully understand how the atmosphere responds and recovers.”

The study’s MIT co-authors include Peidong Wang, Yaowei Li, and Kane Stone; along with Benjamin Santer of the University of East Anglia; Qiang Fu of the University of Washington; Rolando Garcia, Douglas Kinnison, and Jun Zhang of the National Center for Atmospheric Research; Jean-Francois Lamarque of Climate Modeling and Analysis LLC; and Gabriel Chiodo of the Spanish National Research Council. 

Chlorine connection

Ozone is a highly reactive molecule, made from three oxygen atoms, that exists naturally in the upper layers of the atmosphere. In the stratosphere, ozone acts as a shield, absorbing the sun’s rays and reducing the harmful ultraviolet radiation that can reach the Earth’s surface. 

In the late 1980s, after scientists first observed signs of ozone depletion in the Antarctic, Solomon led expeditions to the region to measure the stratosphere’s composition. Those measurements confirmed that ozone’s agent of destruction was CFCs — the chemicals which were used globally in refrigeration, air conditioning, and aerosol propellants, among other uses. 

Specifically, Solomon measured higher-than-expected levels of chlorine dioxide in the Antarctic stratosphere. The presence of this molecule, in the same place where ozone depletion was observed, had only one chemical explanation: Ozone was being broken apart by rogue atoms of chlorine. At the time, chlorine-heavy CFCs were in wide use, and MIT chemist Mario Molina proposed that if CFCs drifted up to the stratosphere, photons from the sun could break apart the molecules and release atoms of chlorine, which would then be free to break apart ozone’s oxygen atoms. 

Molina’s work, and Solomon’s measurements, were key in showing that CFCs could deplete ozone — a discovery that earned Molina a share of the 1995 Nobel Prize in Chemistry. Soon after, nearly every country in the world signed the Montreal Protocol, which ultimately led to the successful phase-out of CFCs and other ozone-depleting substances. In recent years, as a result of that global cooperation, scientists have observed initial signs of ozone recovery.

“We know what we have now, and ozone is starting to recover,” Solomon says. “But no one has ever really documented where and when and why the first ozone depletion would have happened.”

Signal over noise

For their new study, Solomon, Guan, and their colleagues took a “what-if” approach, posing the question: What if the past had the monitoring capabilities of the present? When would we have been able to detect the earliest sign of human-induced ozone depletion? 

Today’s monitoring tools are sensitive to a certain signal to noise, meaning they can identify patterns of ozone loss that are more likely a “signal” of human-induced depletion (such as from CFCs), versus ozone loss that is due to “noise,” such as random fluctuations from weather and natural phenomena. 

With this in mind, the team looked to reproduce the chemistry of the atmosphere over the last century to see whether they could see a signal over the noise, based on the sensitivity of today’s monitoring tools. 

The team used 16 different model runs, each of which simulates varying conditions and dynamics of the atmosphere at various latitudes and altitudes, as well as the concentrations and interactions of ozone and other molecules. Ozone is affected by not only human-caused chemicals but also natural phenomena such as volcanic eruptions and El Niño weather patterns. Each model run simulates ozone’s response to these natural phenomena, which the team combined to establish a range of “noise,” or ozone depletion that likely is due to natural variability.

They added to each model the various industrial chemicals that were known to have been produced at various times over the last century. 

“Year by year, we have estimates from industry of how much of these chemicals were made and sold globally, and the emissions of all these chemicals, which the models include,” Solomon explains. “And in the case of carbon tetrachloride, the really cool thing is, we also have ice core data.”

Ice cores are drilled-out cylinders of deeply buried ice, that had formed in the Antarctic and Arctic from the falling and layering of snow over hundreds of years. Ice cores contain the remnants of snow, as well as whatever trace chemicals in the atmosphere the snow originally fell through. Scientists can therefore use ice cores to estimate the composition of the atmosphere through history. 

“We actually see in the ice cores that carbon tetrachloride starts increasing already by the 1940s,” Solomon notes. 

The team incorporated industrial and ice core data into their models, then looked to see whether a signal of human-induced ozone loss stood out from the noise of natural fluctuations. Their analysis revealed that a signal did appear, as early as 1957. Not only did they see when the signal appeared, but also where: in the tropics, rather than the Antarctic. 

The researchers say that human-induced ozone loss was likely occurring globally, but was easier to spot in the tropical upper stratosphere, since that is the region where the range of natural fluctuations is the smallest, and therefore where a signal can stand out better.

Finally, the analysis indicated that carbon tetrachloride, and not CFCs, was the cause of the earliest ozone depletion. 

“That’s the only ozone-depleting substance that was increasing that early,” Solomon says. “We started using carbon tetrachloride in the 1930s as a dry-cleaning agent, and as a degreasing solvent. We didn’t start using CFCs until quite a bit later.”

Carbon tetrachloride has since been phased out of use in most of the world, initially due to its health concerns; the chemical can cause nervous system disorders with prolonged exposure and is a suspected carcinogen. Since the Montreal Protocol began to tightly limit its use in the 1990s, the molecule’s concentrations in the atmosphere have been on a decline. Still, Solomon says the new study highlights the need for vigilance in monitoring carbon tetrachloride, CFCs, and other ozone-depleting substances that may have been phased out but can still linger for decades.

“We’ve gone through a big effort to get rid of these chemicals,” Solomon says. “Don’t we have an obligation to keep monitoring to make sure the atmosphere responds the way we think it should?”

This research was supported, in part, by the National Science Foundation, the National Oceanic and Atmospheric Administration, and the European Commission.