Paradox (noun, “PAR-uh-dox”)

A paradox is when a scenario seems — at first — impossible. It goes against what your gut feeling tells you is true. Once resolved, however, a paradox often reveals something crucial about how a system works. It may correct a common misconception. Or it may address incorrect assumptions.

Sometimes, a paradox consists of real facts that seem to defy expectation. Moravec’s paradox is one example. It notes that many abilities that come easily to humans are quite hard to program into a robot. Walking, for instance. On the flip side, many easy-to-program abilities, such as complex math, are hard for humans to learn. Exploring this paradox can help us better understand how the human mind works.

Peto’s paradox is another example. It points out that large, long-lived animals get cancer at about the same rate as small, short-lived ones. For a long time, that fact puzzled scientists. They expected large animals to get cancer more often. Cancer starts when something goes wrong with a single cell’s genetic material. Shouldn’t large animals — which contain many more cells — have this problem much more often than small ones? The data showed this was not the case.

Scientists studied this paradox. They learned that large animals have genes that help reduce cancer risks. Researchers did not know to look for such genes before noticing the paradox. This is one case of a paradox helping us learn things we didn’t know that we didn’t know.

Some paradoxes are thought experiments. That’s an imaginary scenario that allows us to explore an idea that we can’t test in real life. Imagine, for instance, you traveled to the distant past. But then, you accidentally killed a distant ancestor. (Oops!) This might mean that you would no longer be born. But wait. If you no longer exist, that means you wouldn’t be able to travel back in time to kill your ancestor in the first place, right?  

On its own, this sci-fi paradox doesn’t answer questions. But it does help us think about the physics of time. And that can inspire new ideas in scientists who study the nature of our universe.

In a sentence

The “paradox of horror” refers to how people often seek out films that seem — on first glance — to evoke only negative emotions.

Check out the full list of Scientists Say.

In the movie WALL-E, the last Earthlings travel through the Kuiper Belt on board the starship Axiom. For 700 years, a fully automated crew of robots has cared for them now that Earth has become inhospitable. Running the ship’s operations is AUTO. AUTO is an artificial intelligence (AI) system working to keep humans away from Earth — forever.  

Here at home, space agencies like NASA and the European Space Agency are using AI to explore our solar system. This includes guiding rovers on Mars, preventing satellite collisions and training astronauts for spaceflight. But these AI systems are much more limited than their fictional counterparts.  

Today’s AI is much more unpredictable and prone to mistakes than what you see in fiction, says Daniele Gammelli. At Stanford University in California, he studies how to integrate AI systems into robots that interact with their environment.  

Anyone using ChatGPT has probably seen how it hallucinates, or makes up inaccurate information. AI systems in space robots would need to complete multi-step tasks in all sorts of scenarios without mistakes. After all, Gammelli says, in space, “you have virtually no room for error.”  

The little robot that could 

The title robot WALL-E is a trash-compacting machine — one that abandons his duties to follow another robot, EVE. WALL-E’s greatest strength is, arguably, his ability to handle change. To escape a self-destructing pod, WALL-E uses a fire extinguisher to flee. When his wheel or eye malfunctions, WALL-E can replace the damaged part. All this is learned from experience and done without additional programming.  

Such versatility is an example of artificial general intelligence (AGI). This type of AI doesn’t yet exist. It would think and learn like a human across different situations. AGI could also take on tasks for which it had not been programmed. 

Adapting to unforeseen situations is a big goal for future space robots, says Gammelli. Between radiation, extreme temperatures and space debris, space is a dangerous, ever-changing environment. Any robot navigating space would need to handle evolving conditions — many of which don’t exist on Earth.  

“The kind of scenarios you are kind of forcing on your robot are, by definition, things that nobody has ever seen,” he says. 

Today’s AI excels at single tasks or several closely related ones. Known as narrow AI, it handles repetitive and predictable work well. “I’d say the top skill that it excels in is processing a huge amount of data very efficiently,” says Sanjoy Paul in Houston, Texas. There, at Rice University, he studies how AI can assist in space missions. 

Martian rovers use narrow AI. To select rock samples, Perseverance employs AI algorithms to scan minerals and assess if they’re worth collecting. Another rover, Curiosity, uses AI to zap rocks with lasers based on their shape and color. All of this is done without human input.  

Any human sorting through that kind of data could get overwhelmed, says Paul. When faced with huge amounts of data, people can struggle to make decisions. Why? Picking out important details becomes much harder. “AI can actually cut through all the details … and highlight those things for humans to take a look at,” he says. 

To handle multi-step tasks, nearly all space robots rely on so-called autonomy stacks, explains Gammelli. These are separate, but linked, modules that tackle different actions. For instance, one AI model might use cameras or sensors to detect rocks or obstacles. This info would be passed to another module. It would interpret those results and determine the appropriate next action. Other modules would then carry out the physical maneuvers to get that job done.  

Still, today’s robots fall short of what people can physically do, says Paul. Many of the basic actions we perform require lots of coordination. Researchers are still studying how to design AI models to carry out actions as efficiently as possible, says Paul.  

Would you trust AI with your life? 

Aboard the fictional Axiom, robots manage everything. Custodial robots scrub and polish. Utility bots handle repairs and maintenance. Hover chairs cart the ship’s residents to their destinations. This lets Axiom’s passengers live sedentary lives, spending their days watching videos and drinking food shakes.  

People would probably need to play a bigger role in space travel than what was depicted in WALL-E, says Paul. While AI continues to improve by leaps and bounds, it remains unpredictable. With lives at risk, one small mistake could prove disastrous. “If your life depends on it, would you really bank on AI? Probably not,” he says. “You still need humans in the loop.” 

Most likely, he suspects, space travelers would rely on AI as an assistant. That might look more like another fictional AI system: J.A.R.V.I.S. from the Iron Man series. Tasked with monitoring Tony Stark’s buildings and power suits, J.A.R.V.I.S. acts more like a team member than a captain. It synthesizes data and offers suggestions to Stark, who remains in control. With large language models like ChatGPT and Claude, that option seems much more feasible, says Paul.  

Still, people wouldn’t want to micromanage all spacecraft, says Gammelli. Machines like rovers should eventually be able to make their own mini-goals, he says. These would all align with the engineer’s overall mission.  

Such mini-goals would allow bots to handle unforeseen situations better. It would also free up people to take care of more critical tasks and decisions. “We want these robots to be as independent as possible,” he says. Though maybe not as independent as mission-quitting WALL-E. 

The three-banded armadillo has an unusual way to protect itself. When threatened, it rolls itself up. The adorable mammal’s thick plates form a nearly impenetrable ball that just may save its life. That ability has now inspired the outer shell for a new robot — one that can keep fragile stuff, such as electronics, from getting damaged.

Like an armadillo, the new device changes shape to protect itself with a rigid shell, says study leader Jianyu Zhou. He’s a mechanical engineer at North Carolina State University, in Raleigh.

His team’s new gadget starts out flat. But when touched, it can roll up into a wheel. A hard, 3-D printed outer surface will guard its contents from puncture, breakage or falls.

The design can protect components built into a device. But there’s also “an inner space,” Zhou adds. So it can hold and protect a small payload, too.

His NC State team describes its morphing robot and how it might be used in the May 27 Science Advances.

How to balance rigid and flexible

Zhou works in the lab of Yong Zhu, much of whose work has focused on flexible electronic devices — ones whose electronic innards tend to be fragile. Such gadgets might be wearable fitness trackers or patches that measure blood pressure or blood-sugar levels.

Zhu’s group also has created soft, squishy robots. Inspired by caterpillars, these might wriggle through debris after a natural disaster to help search for survivors or assess damage. They also might be sent out to explore and study hazardous sites, such as the floor of the deep ocean or areas near an active volcano.

Such outings could put soft, morphing robots at risk of getting crushed, fried or pulverized. “The natural question,” Zhu says, has been “how to protect those.”

When he and Zhou set out to explore new defenses for fragile electronics a few years ago, they looked to nature.

Some animals, such as turtles, have hard shells to keep predators or other dangers at bay. But the armor protecting three-banded armadillos and pangolins has another benefit: It can change shape, allowing the animals to form an almost perfect sphere.

Despite being able to morph, it offers a protective “rigid shell,” Zhou points out.

Anatomy of the robo-armadillo

It took the engineers more than a year to create their new robot. The challenge: They had to find a design that could curl up and lock into itself. For this, they took a layered approach. On the outside is a hard shell made from rigid, protective plates. They not only connect but also bend.

Inside this shell are layers of a flexible material that acts like a synthetic muscle. One layer expands when heated. Under that is a layer that contracts when heated. The innermost layer of the device is an inner skeleton made of almost-triangular pieces that lock together as the device rolls up. Zhou says it comes together like an origami creation.

Next, the team had to create a sensor that, when triggered, will set the shell in motion. When it detects a vibration nearby, which could be as gentle as a touch, this sensor sends a signal to an actuator. That component turns on a tiny heater that warms the muscle-like material. That material contracts, pulling everything together to form a wheel-like structure. The entire process takes about a minute.

View this post on Instagram

Nina Attias has been studying armadillos for decades as a conservation biologist in Brazil. She works for the Wildlife Conservation Network, based in San Francisco, Calif. “I am passionate about armadillos, I love them,” she says. She’s happy, but not surprised, that these animals inspired the new robot. “Once you start knowing about them,” she says, “you see that these are pretty amazing creatures.”

The new device impressively mimics the armadillo’s strengths, she says. Only three of the 20 known armadillo species can roll into a ball. Those include the three-banded armadillo native to South America. It’s super-sensitive to vibrations, she adds. It can sense yummy underground termites. These armadillos also can detect the approaching steps of a jaguar that views them as lunch.

After sensing a touch, the new device similarly rolls itself up. And its contracting material, between the shell and skeleton, act like the strong muscles that keep an armadillo’s armor in place. When an armadillo rolls up, “you can’t open it for anything,” Attias says.

The new device survived a range of tests. For instance, the researchers dropped it from a few feet onto a bed of tacks. Fragile objects inside it, including glass spheres and raspberries, remained unharmed.

What’s exciting about the new device, Zhou says, is how it brings rigid and soft parts together. “The armadillo offers a good balance between function and protection,” he says.

Zhu says it’s too soon to know how people might decide to use such a device. But he’s encouraged by the success of a nature-inspired way to solve an engineering problem. In future work, he says that he’ll be looking at other animals for more inspiration.

Nearly 2 million years ago, a leopard in Africa sank its teeth into the face of an early human. This bite etched tiny tooth marks into the unlucky person’s jawbone. That bone now speaks to a chilling early chapter in our history: when humans were prey.

The fossil belongs to Homo habilis. It’s one of the earliest members of our genus. (Homo includes modern humans and our closest relatives.) At first glance, the bone marks look like scratches. But a recent, close analysis tells another story. The bite marks suggest leopards hunted these early folk.

That discovery helps solve a fossil mystery. It also raises questions about a time when humans held a more precarious spot in the food web. Sometimes people ate meat. Other times, they were the meat being eaten.

That threat of becoming prey to some other predator likely shaped how humans evolved. To survive, our ancestors learned to count on each other for protection. They may have lived and worked in groups, sharing information about dangers and teaching each other skills. Over generations, these behaviors could have helped people become more cooperative and successful.

Today, it’s easy to imagine our early ancestors as fearless hunters, chasing prey across ancient savannas. But that image reflects a time late in our evolution. Before we became one of the world’s most successful predators, we first had to survive life as prey.

When humans were on the menu

Much of what scientists have learned about ancient humans comes from fossils found at Olduvai Gorge, in Tanzania. Since the 1930s, researchers have unearthed many bones from our early ancestors at this site in southeastern Africa. They include a pair of tooth-marked jawbones.

Other finds include tools made of stone or bone.

Sharp stones likely served as knives. Cut marks on animal bones suggest early humans ate meat. This led many scientists to assume H. habilis was a hunter. But these early people may have dined on leftovers scavenged from bigger, stronger predators.

The bitten human jawbones add another angle to this story.

Marina Vegara-Riquelme studies our human ancestors and their fossils at the University of Alcalá in Spain. She led a team that examined H. habilis bones using artificial intelligence, or AI. Her group wanted to know what animal had left its bite marks on a human jaw.

Different carnivores have different eating habits. Hyenas often scavenge dead carcasses. Leopards, in contrast, rarely eat meat they didn’t kill. Knowing which one left the marks would suggest whether the H. habilis had been hunted or died some other way.

Even experts can struggle to tell apart the tooth marks of a leopard from those of a hyena, Vegara-Riquelme notes. But AI can. By comparing thousands of details, it can spot patterns too subtle for the human eye to see. Her team used thousands of tooth-mark images from modern carnivores to train the AI. Then they showed it the bones in question.

“Two Homo habilis specimens show tooth marks most likely inflicted by leopards,” Vegara-Riquelme says. That makes it look less like scavenging.  

Study co-author Manuel Domínguez-Rodrigo co-directs the Institute of Evolution in Africa in Madrid, Spain. To him, it’s clear how the encounter ended. “That individual was preyed upon,” he says. “Not scavenged.”  

If true, what had the lives of H. habilis been like with lions and leopards as neighbors? Could survival have required new traits that helped our ancestors ascend the food web?

Life in a world of predators

For millions of years before H. habilis, our ape ancestors spent most of their time in trees. They might venture to the ground now and then. But when danger emerged, they’d dart into the trees and out of reach.

By the time H. habilis evolved, humans were walking upright (though not as well as today’s H. sapiens), spending far more time on the ground.

An open savanna offered new opportunities. Ambling upright helped early humans travel farther. It also gave them a better vantage to spot potential food across the landscape. But being out in the open may have made them easier targets, too, for powerful predators.

Craig Stanford studies predator and prey relationships among primates. He’s an evolutionary biologist at the University of Southern California in Los Angeles. He’s spent years in Tanzania watching Africa’s monkeys work together and protect one another.

Life on the open savanna would have been risky for early humans, he says. Even if predators killed only a few people, the threat would have shaped how early species lived and behaved.

“Predation may have been rare, but that doesn’t mean it was unimportant,” Stanford says. “There’s no question that predation influenced our evolution.”

Many prey species rely on group living for protection. “You have more eyes to detect a predator coming,” Stanford explains. When one animal spots danger and yells an alarm, others can run or climb to safety.

And safety in numbers isn’t just about watching. It can also help to fight back.

In Tanzania, Stanford saw male colobus monkeys work together to fend off threats. They would leap onto chimpanzees that threatened them. An adult male chimp weighs about 54 kilograms (120 pounds); a colobus monkey only about 9 kg (20 lbs). “So three [monkeys] together weigh half of what the chimp weighs,” Stanford says. “They’re biting the chimps.” Such group attacks could fend off a chimp.

Early humans on the open savanna may have similarly benefited from teaming up. Faced with threats from big cats, people who were more alert and willing to cooperate may have been more likely to survive. If so, they could have passed along those traits to their offspring.

Working together for a better dinner

Survival wasn’t just about not becoming someone else’s meal, though. It also was about getting access to your own high-quality food. Here, too, cooperation would have helped.

Briana Pobiner works at the Smithsonian’s National Museum of Natural History in Washington, D.C. She studies what ancient humans ate and how they got their food. Marks on fossil bones can say a lot about this, she says.

Early humans would have cut meat from wherever they could get it. Consider butchery marks on bones that don’t have a lot of meat on them — “like ankle bones and lower foot and leg bones,” says Pobiner. Here, she says, “We think [humans] mostly got the scraps after the predators ate.”

Such a pattern suggests people were scavenging.

With large prey, early humans likely ate after big carnivores had already taken the better meat. That doesn’t mean these humans didn’t hunt. They may have captured smaller animals on their own. Taking down large prey, however, would have been much harder and more dangerous.

“Early humans probably got the not-great parts of animals when they were able to eat meat,” Pobiner says. “The predators were getting the best parts.”

But, she adds, “At some point that changed.” By around 1.5 million years ago, fossils show stone-tool cut marks on bones from some of the meatiest parts of big game. By now, it seems, people were no longer eating just the leftovers.

Fossils can’t directly show cooperation. But the bone markings suggest humans gained better access to meat over time. Researchers think teamwork may have helped make that possible.

Working in groups would have given humans a better chance of getting larger prey and better meat. Sharing meat may have strengthened social bonds. Plus, large carcasses are hard for one person to process or defend.

Such cooperation may have paved the way for humans to become more successful.

Learning to control fire

People likely evolved group living for several reasons, cautions Richard Wrangham. He works at Harvard University in Cambridge, Mass. As a biological anthropologist, he studies how fire has shaped human evolution.

Around 1.8 million to 1.5 million years ago, fossils show the rise of a new species of human, Homo erectus. Compared to earlier species, such as H. habilis, H. erectus had longer legs, a larger body and a bigger brain. These changes suggest humans were better adapting to life on the ground. This may have included sleeping there rather than in trees.

Here, fire also may have been key. “Once you’ve got fire, it increases the amount of protection against predators substantially,” Wrangham says. Flames and smoke would have scared off big cats, especially at night. He says, “I would imagine that there was much more predation of [early humans] prior to fire.”

Groups able to control fire may no longer have needed to retreat to trees each night. Sleeping on the ground would have marked a major shift in human evolution. With time, arms got shorter. Legs got longer. Walking and running became more efficient. Such changes would have made them more effective hunters, too.

People who could travel long distances could also track prey across broad savannas.

Fire had another benefit: It helped people get more energy from their food.

Cooked food is easier to digest. That extra energy could have allowed brains to grow bigger and more complex. Wrangham explored this idea in the 2010 book, Catching Fire: How Cooking Made Us Human.

Ours is the only human species alive today. But earlier in history, multiple human species lived — and evolved — beside each other. Fossils suggest that not all of these species advanced at the same pace. The pressures of avoiding predators while getting enough food may have nudged some species higher in the food web. As a result of these gradual changes, some species flourished. All but one of them — H. sapiens — died out.

By 300,000 years ago, stone-tipped spears appeared. Archaeologists found dozens of these along with the bones of hunted and butchered animals at a site in Germany. The find suggests that people had become skilled hunters. They planned ahead and worked in groups to hunt big animals. No longer just surviving within their ecosystem, humans were beginning to dominate it.

Those leopard bites on ancient jawbones remind us that humans haven’t always been on top. Our ancestors came to survive by working together, sharing resources and developing new tools and skills. Over time, these behaviors helped people shift from fearing predators to competing with them.

That almost nothing hunts us is a fairly recent development. Our history has been shaped by our vulnerability and the behaviors that allowed our ancestors to survive long enough to thrive.  

While at Brown University in Providence, R.I., oceanographer Lily Dove studied the physics of Antarctica’s Southern Ocean with the help of some unusual, furry assistants: Antarctic seals. “They’re oceanographers,” she says. “They are regularly making dives in areas that we don’t often get to sample.”

Dove didn’t work with the seals directly. Instead, she interpreted data collected by the Marine Mammals Exploring the Oceans Pole to Pole program. This international group of scientists collects ocean data through sensors attached to the foreheads of southern elephant and Weddell seals. (Don’t worry. The animals don’t mind.) The seals dive more than 100 meters (330 feet) in search of food, allowing researchers to monitor temperature, salinity and pressure at depths inaccessible to human divers.

Dove used these data to study how climate change is impacting the Southern Ocean. That’s the vast ocean that circles Antarctica. In certain areas of the Southern Ocean, atmospheric heat moves from the surface to the deep ocean. This heat is typically trapped for hundreds of years once it reaches the depths, further increasing ocean temperatures.

Now at the Georgia Institute of Technology, Dove works with autonomous robots to study polar oceans. Some, such as floats, ride sea currents. Others, such as gliders, are pilotable underwater vehicles, meaning Dove can target specific locations. “And this is all remote,” she says. “I can sit here in Georgia and tell my glider in a whole different hemisphere where I want it to go.” 

This tech is also more predictable than her past assistants. “Seals are very independent,” she says. Unlike a glider, “seals very much tell me where they want to go,” she says. In this interview, Dove shares her experience and advice with Science News Explores. (This interview has been edited for content and readability.)

What inspired your career? 

Growing up in Florida, the ocean interested me as this powerful system that we take for granted. But I didn’t really consider it as a career back then. I knew I loved science, like physics and chemistry. In college, I knew I wanted to do one of those things or some combination. I changed majors like five times.

Then I realized that I liked those topics because they could help explain what I cared about: the Earth and the environment. I learned that you need all different kinds of science to understand the Earth system. Biology, physics, chemistry, geology — all of them. I ended up researching atmospheric chemistry.

How did you get to where you are today? 

In graduate school, I took those skills from studying atmospheric chemistry and started doing oceanography instead. The atmosphere and ocean have the same physics. That made it feel like it was the perfect transition for me. I landed in a group that was investigating polar oceans, which wasn’t something I’d even considered before. I went to the Southern Ocean for the first time in 2020. The polar oceans are pretty remote and far away from where most people live. But they have massive impacts on our climate system, especially the Southern Ocean. It wasn’t until I was there that I realized that. That’s when I knew that this is exactly what I wanted to be doing.

What was your biggest challenge, and how did you overcome it? 

I’ve experienced quite a bit of discrimination in this field as a woman. There are a lot of assumptions people have about who is welcome to do polar science. Historically, women have not been included on research cruises. Women used to stow away on ships to get data for their research theses. There’s a famous case at the Woods Hole Oceanographic Institution of a woman stowing away on a research vessel in the 1950s.

That stigma is still present to an extent, but it’s changing a lot. And it’s not just about women. People of color, queer people and disabled individuals are also experiencing this less-overt discrimination.

Throughout my time in the field and in academia, there have been times when people have assumed I can’t do something. That’s really frustrating, but it’s also given me the skill to really speak up and advocate for myself.

What advice would you give to your younger self? 

Say “yes” to as many opportunities as possible. That could be attending an event, talking with a scientist or taking a course. The worst-case scenario is that you learn something new that’s just not super interesting to you. The best-case scenario is that something grabs your attention, and you can’t let it go.

I am also glad that I didn’t hyperfixate on a particular area of study super early. I’m glad I got a strong basis in a lot of different scientific fields before I decided on a focus. It allowed me to be really flexible and open to new ideas.

If you already know that you’d want to do something specific, listen to your gut. But if you are not quite sure, lean into just learning a lot.

When you hear the word “fossil,” you might at first think of a dinosaur skeleton at your local museum. Or perhaps the imprints of leaves or shells in a rock from a nearby creek. Both are examples of body fossils. But there is a second type of fossil: trace fossils.

Body and trace fossils are created through similar processes. In both cases, some trace of an ancient life form is left in sediment. That remnant gets preserved when the sediment transforms into hard rock. But while body fossils preserve the remains of ancient life forms themselves, trace fossils keep a record of those life forms’ activities. Examples include burrows, nests, footprints, poop and vomit.

Ichnologists use trace fossils to obtain snapshots of ancient life forms’ daily lives that these scientists can’t get from remains alone. Take, for instance, fossilized poop, called a coprolite. Coprolites can offer clues about an ancient creature’s diet.

Fossil dung has revealed that, more than 150 million years ago, flying reptiles called pterosaurs ate mollusks and crustaceans. These fossils have also shown that tens of thousands of years ago, squirrels may have snacked on the remains of woolly mammoths and wolves.

Dietary info can also come from fossil vomit, known as regurgitalite, and fossilized stomach contents, called cololites.

The marks that ancient animals carved into their environments can also make useful trace fossils. Rocks from coastal Taiwan, for example, contain burrows likely dug by giant worms some 20 million years ago. Those lairs suggest that the worms lurked beneath the seafloor like Star Wars monsters and shot out to attack their prey.

Scientists have also studied dinosaurs’ footprints to tease out details like how fast they ran.

Want to know more? We’ve got some stories to get you started:

New clues about dino speed come from birds strutting through mud Scientists use dinosaurs’ tracks in wet ground to estimate how fast they ran millions of years ago. (8/22/2025) Readability: 6.4

Frozen poop reveals Ice Age squirrels’ ecosystem DNA stored in ancient scat shows the Yukon rodents ate their way through the food chain. (8/25/2026) Readability: 6.8

Giant worms may have hidden beneath the ancient seafloor to ambush prey Trace fossils may be evidence of predatory behavior similar to modern bobbit worms. (2/16/2021) Readability: 8.1

Explore more

Scientists Say: Ichnology

Scientists Say: Coprolite

Explainer: How a fossil forms

Fossil vomit shows what one 290-million-year-old predator dined on

New fossils bring the wide world of pterosaurs to life

Fossil stomach reveals a dinosaur’s last meal

New beetle species found in fossil poop of this dino relative

A love of small mammals drives this scientist

The secret to T. rex‘s incredible biting force is at last revealed

Activities

Word find

Want to see how easy it is for ancient life forms to leave traces behind? Make your own trace fossil with instructions from the Natural History Museum of Los Angeles County.

Proteins are made of building blocks known as amino acids. Human proteins are made of various combinations of 20 of these. A diet especially rich in one of them — arginine — appears to help the immune system. It can fend off infections and at least one type of cancer, finds a new study in mice.

Arginine appears to ramp up the immune system’s ability to spot danger. A high-arginine diet seemed to offer the rodents some protection against flu and COVID-19 viruses. It also quashed colon tumors, says Sohail Tavazoie. A doctor and cancer biologist, Tavazoie works at Rockefeller University in New York City.

His team shared the new findings July 30 in Cell.

“I don’t want the takeaway to be that everyone should be just consuming arginine all the time,” says Tavazoie. It’s just too soon, he says, to know if taking it as a food supplement could prevent or even treat these diseases in people. But he’s interested in launching human studies to test that.

The new data do make a compelling case for further testing, says Costas Lyssiotis. He had no role in the new study. But he is a cancer researcher at the University of Michigan Medical School in Ann Arbor.

Might a high-arginine diet help the human immune system recognize tumors and target them for destruction? That’s not yet clear, Lyssiotis says. But, he adds, it’s something that is “super provocative and should be pursued.”

Not ‘essential’ — but beneficial

Research has been suggesting that arginine may play an outsized role in disease.

Nine amino acids are considered “essential” nutrients. That’s because our bodies use them but can’t make them; they must be acquired as part of what we eat. Arginine is not one of those. But because this amino acid can open blood vessels, some people do take arginine as a food supplement. The goal is to help their blood circulate.

Arginine also tends to be scanty in the lungs of people with influenza or COVID-19. It’s barely present, too, in the tumors that develop in people with colon cancer. What’s more, studies have shown that both cancer and viral infections can crank up the destruction of arginine.

Tavazoie’s team wondered what was so important about this amino acid that multiple diseases shut it down.

In the lab, his group tinkered with arginine levels in mouse and human cells. Cutting back on arginine reduced levels of key immune proteins, they found.

Take what’s known as MHC class 1 proteins. Riding on the surface of our cells, they tell the immune system what to watch out for. Like a police officer holding up a sketch of a criminal, they display snippets of viral or cancer proteins. But the immune system has trouble recognizing and rousting those outlaws when cells have fewer MHC-1 proteins than normal. 

A key player in immunity?

Tavazoie’s team then turned to mice at risk of colon cancer. When the researchers fed them a low-arginine diet, MHC-1 levels fell and the number of colon tumors rose. But add bonus arginine to the diet — what would be about 10 grams a day for humans — and the outcome flipped. Now MHC-1 levels rose and tumor-forming fell.

These data suggest dietary arginine can control MHC-1, a very important group of immune proteins, Tavazoie says.

Arginine-starved cells slowed their assembly of MHC-1 proteins. That may be because these proteins contain long stretches of arginine units. If there aren’t enough of these building blocks available, “there’s a supply-chain shortage,” he says. At that point, MHC-1 construction stalls.

The team observed a similar story in mice infected with the flu virus or the virus that causes COVID-19. Low-arginine diets led to them having low levels of MHC-1 proteins. They also had more viral proteins and inflammation in their lungs compared to mice eating high-arginine diets.

This suggests arginine was blunting the effects of infection. And arginine supplements given after flu infection boosted mouse survival, the team found. That points to it as a possible treatment one day. Tavazoie now wonders whether other amino acids may play similarly important roles.

There’s still much work needed to know if the mouse data predict effects in people, says Pierre Close. He’s a cancer researcher at the University of Liège in Belgium. He didn’t participate in the new study. But, he says, this research might one day lead to therapies that help the immune system better tackle disease. “We are all looking for new strategies.”

Antimatter (noun, “AN-tee-MAT-ter”)

Antimatter is made of antiparticles. Think of antiparticles like partners of ordinary particles, but with the opposite properties.

You’ve likely heard of protons, neutrons and electrons. These are the tiny particles that make up atoms. But you may not have heard of their anti-twins. These are antiprotons, antineutrons and positrons.

Particles and antiparticles have the same mass. But they generally have opposite electric charges. An electron is negative. So the positron is positive. A proton is positive. So the antiproton is negative. But exceptions to law of opposites exist. Take the neutron, which has no electric charge. It’s neutral. So the antineutron is also neutral.

When matter and antimatter meet, they destroy each other. If an electron meets a positron, for instance, they self-destruct. As physicists say, they “annihilate” each other. This releases energy in the form of gamma rays. That’s a high-energy type of light.

Antiparticles can bind together to form atoms, just as ordinary particles do. Scientists have built antihydrogen in the lab. They do that by pairing a positron and an antiproton. This isn’t an easy feat, though. If either of those antiparticles touch ordinary matter — poof! They cease to exist. That means scientists must “handle” the particles without touching them. They do this through magnetic or electrical fields.

Very, very little antimatter exists today. A tiny amount occurs naturally. For example, cosmic rays constantly smash into Earth’s atmosphere. When they do, they produce tiny bits of antimatter. Some radioactive decay gives off antiparticles, too. But antimatter never lasts very long. Normal matter quickly destroys it.

Scientists believe that the Big Bang produced equal amounts of matter and antimatter. Why does ordinary matter reign supreme today? That’s one of the universe’s biggest mysteries.

In a sentence

Scientists search for mysterious forms of matter such as dark matter and antimatter hoping to reveal secrets of the Big Bang.

Check out the full list of Scientists Say.

Our sweets cravings may have set the stage for the human brain to become big and powerful. That’s the finding of a new study.

The brains of our upright-walking ancestors, some 4 million years ago, weighed only around 300 grams (11 ounces). That’s about equal to a medium apple. Today, the human brain averages 1,500 grams (3.3 pounds). That’s about equal to 1.5 liters of water. Much of this brain expansion took place before early humans mastered both fire and cooking. That’s a point that marked when they could start regularly dining on energy-dense starches.

Carbohydrates come in two main classes. Sugars tend to be relatively simple molecules. Starches tend to be bigger and more complex. A new analysis of existing data from human ancestors, as well as chimps, suggests that a large share of their energy probably came from sugary carbs, such as those in fruits and honey.

Simple sugars may have played an overlooked — but important — role in human evolution, researchers now conclude. They describe why in the August 6 Science.

Until now, much of how diet affected human evolution has centered on meat. Around 2.5 million years ago, our early upright-walking ancestors began eating more meat — raw at first, then eventually cooked. Meat is rich in protein and fat. And that appeared to play a big role in fueling our ancestors’ ever-expanding brains.

But Jennie Brand-Miller thought there might be more to the story. A nutrition scientist at the University of Sydney, in Australia, she wondered how sugars might have affected the evolution of our brains.

A 2017 study had shown that fruit-eating primates had bigger brains than leaf-eating ones. Eating fruit boosts demands on various thought processes, other researchers had suggested two decades earlier. Might this have kickstarted the evolution of big brains in humans?

Explains Brand-Miller, “You need to remember when various species of ripe fruit begin to ripen, then you need to remember where it is in the forest.” And, she adds, “Your memory is associated with a bigger brain.”

Fossil primates and later hominids, such as Homo erectus, showed signs of tooth decay, Brand-Miller notes. This suggests a diet containing soft, sugary foods. (Homo erectus was the first of our ancient cousins to venture outside Africa; that was 2 million years ago).

“All these things kept adding pieces to this lovely tapestry, this puzzle that said fruit-containing sugars must be important [in human evolution],” she says.

Still, what about meat?

Brand-Miller is part of a team that has now calculated the dietary needs of chimps, humans and three early species of human ancestors. To do this, the researchers combined data from fossils, teeth, the species’ metabolism and the biology of bodily processes.

Our earliest chimp-like ancestors from around 4 million years ago were largely vegetarians, they found. Fruit likely provided more than two-thirds of their food energy. And it played a big role as hominid brains grew over the next few million years, the data also suggest.

Clever foraging skills would have helped fuel their energy-sapping brains. These would have included hunting, cracking nuts, grinding seeds or extracting bone marrow. Later, once cooking became common, humans could have turned to starches (such as grains, beans and tubers). But a constant need for sugary foods appears to have been a key player in our evolution. And it didn’t disappear after a shift to foods high in proteins and fat, such as meats.

Meat and sugars? “I see them as parallel stories, not one replacing the other,” Brand-Miller now says. “Animal foods were clearly important. They’re very nutrient dense.”

A past focus on meat-eating may have missed some food trends that helped build the human brain, agrees Julie Lee-Thorp. She’s an archaeologist at the University of Oxford in England. She did not take part in the new study.

Focusing on meat’s role may reflect a modern bias, she adds. “Meat is a high-status food today,” notes Lee-Thorp. “And that has perhaps been carried over into the way we think about the past.”

But Brian Wood largely dismisses that. The study of human evolution has not been obsessed with meat-eating, he says. A biological anthropologist, Wood works at the University of California, Los Angeles. A diverse range of foods fueled our evolution, he says. And they were made possible by our growing “ability to capture and harvest high-quality, difficult-to-acquire foods.” This included hunting, sure. But, he adds, foraging for fruit, honey, nuts and the underground parts of plants was also part of it.

Earth has frozen over in the past — at least five times. Science fiction is full of takes about how it might happen again and the implications of a modern ice age. The problem: At a minimum, sci-fi gets the timetable for Earth turning into a “snowball” wrong — often laughably.

Here’s how some stories have portrayed it: Earth freezes. Humankind’s remnants huddle in ice caves, fending off giant alien monsters (Snowball Earth). Or people shelter in the New York Public Library as ice encases the Empire State Building (The Day After Tomorrow). Or a train hurtles in an endless loop, both protection and prison for Earth’s scant survivors (Snowpiercer).

These stories are all set in Earth’s near future, within a decade from now. And what makes their visions so chilling is that there’s at least a snowflake of truth in each. (Well, space monsters haven’t invaded Earth — yet.)

The most severe of Earth’s ice ages spanned from 720 million to 635 million years ago. It’s aptly been named the Cryogenian Period, from the Greek krýos (meaning cold) and génesis (meaning birth). At least twice during this period, glaciers extended from pole to pole. It could have amounted to a true Snowball Earth.

Although “maybe it was more of a Slushball Earth,” says Thomas Algeo. A geoscientist, he works at the University of Cincinnati in Ohio. Fossils, he says, suggest that even in the coldest times, seafloor-dwelling algae survived. They would have needed sunlight. So patches of open water must have persisted near the equator, he says.

All such past deep freezes set in due to a series of geologic events. Take the Cryogenian. It developed after shifting tectonic plates caused a supercontinent to split apart. That increased rainfall and rock weathering. These dramatically reduced atmospheric carbon dioxide, which triggered a runaway cooling effect. Widespread ice, which reflects the sun’s warming rays back into space, would have exaggerated that cooling.

All of these processes are slow, which is not ideal for action flicks. Fortunately, sci-fi can play with time. It’s not unusual for fiction to take natural or human-caused climate interactions (and feedback effects) and accelerate or amp them up for drama.

The Day After Tomorrow, for example, which hit theaters in 2004, is based on a real-world scenario: the potential shutdown of a part of a global system of ocean currents. These move heat around the planet. Known as AMOC, this system is short for the Atlantic Meridional Overturning Circulation. Today, AMOC brings relative warmth to regions that might otherwise be frigid.

Back in the mid-1980s, geochemist Wallace Broecker suggested that large releases of freshwater to the North Atlantic Ocean could halt the AMOC. Such slowdowns or shutdowns in ocean currents have played a role in past ice ages, he said.

Scientists today fear it could happen again — and soon.

But to geologists, “soon” doesn’t mean in days to weeks (as portrayed in The Day After Tomorrow). For them, it means perhaps decades to a century.

The Day After Tomorrow’s filmmakers were “quite up-front” about their process, says Stefan Rahmstorf. For ultimate drama, they just “took known weather extremes — tornadoes, storm surges, cyclones, hail storms and blizzards — and amplified those.” Rahmstorf is a physical oceanographer at the Potsdam Institute for Climate Impact Research in Germany. He fact-checked aspects of the film.

Despite its super-short timeline for turning the Northern Hemisphere into a deep freeze, this film did point to some very real science, he notes. For instance, the movie starts out with its science hero giving a talk at a United Nations science meeting. Its topic was “the possible risk of a shutdown of the North Atlantic Current. I gave a very similar talk at such a UN conference in Buenos Aires in 1998,” Rahmstorf notes. “I even showed the same diagram [as the movie’s scientist].”

Snowpiercer imagines a different deep-freeze scenario. Geoengineering goes haywire in this 2013 movie, 2014 graphic novel and TV series (which ran from 2020 to 2024). Here, scientists release aerosols into the sky to reflect sunlight back to space. Their goal was to slow global warming. This idea is, in fact, based on a real type of Earth engineering known as solar-radiation management.

But in this story line, the aerosols proved an overkill. They plunged Earth into a deep freeze within weeks (hours, actually, in the novel).

Geoengineering strategies could cool Earth — just not that quickly. They would more likely take at least a few years. And they wouldn’t be that potent, either. To bring on an ice age, aerosols would have to be deliberately and repeatedly injected over generations, says Douglas MacMartin. He’s a climate engineer at Cornell University in Ithaca, N.Y.

How Earth freezes in Snowball Earth is still something of a mystery, as the manga is still in progress (though a streaming anime version debuted in April). Here, the transition to an ice world takes about a decade. And — spoiler alert — giant alien monsters might have had something to do with it. (For that, scientists don’t really have an answer.)

But the premise behind these stories — that Earth could freeze over — is grounded in very real science. It’s happened before and could again. But it won’t happen quite the way these stories tell it.

One of Earth’s most vital ocean currents is rapidly weakening. The cause: global warming. A collapse of this current could send devastating ripples through the planet’s climate.

The current is known as the Atlantic Meridional Overturning Circulation. Most scientists just call it AMOC. If it were to shut down, the consequences could be catastrophic and far-ranging. In the worst case: Earth’s warming climate could flip to a deep-freeze. Scientists refer to this scenario as a “snowball Earth.”

Why might that happen?

AMOC moves warm surface waters in the Atlantic Ocean from the tropics northward toward Europe. This water ferries with it a massive amount of heat. If that transport were to shut down, scientists say, the world might see a host of dramatic weather. Europe could experience more temperature extremes, of both chilling and heating. The Southern Hemisphere would warm even more than it is now. The North Atlantic would trigger stronger storms. The drought in Africa’s Sahel would worsen. And summer rains in Europe could sharply decrease, with devastating outcomes for farmers.

Researchers have rung warning bells about the AMOC for decades. Its possible shutdown even was the theme of The Day After Tomorrow, a 2004 snowball-Earth disaster movie. 

But predictions of how long AMOC will hold out have been all over the map. And it’s only in the last couple of years that scientists have been able to figure out which way the current is really trending.

Their consensus: It’s not good — AMOC is weakening.

The stakes are so alarming that some scientists have proposed engineering changes to AMOC. One is building a series of dams to seal off the Bering Strait between Russia and Alaska. By cutting off the northward flow of water from the Strait into the AMOC, the current’s strength might stabilize. At least that’s what researchers claim in a report April 24 in Science Advances.

Such proposed changes are controversial. No one can be sure what unknown impacts they might trigger. And the Science Advances paper “is very much a conceptual study,” says Jelle Soons of Utrecht University in the Netherlands. By that he means it’s a clever idea — but no one yet knows how it might be achieved. And Soons should know. He’s not only a physical oceanographer but also a coauthor of that new report.

What such suggestions can do is focus attention on how serious a threat an AMOC shutdown is — and how little time remains to address it.

The role of sea salt

A system of currents — including the AMOC — moves ocean waters around the planet. The water’s density is what drives them.

Saltier masses of water sink below less-salty ones. Colder waters also sink relative to warmer ones. The constant shifting between these masses keeps those currents on the move, almost as if they were riding a conveyor belt.

As the AMOC releases heat to the atmosphere at its North Atlantic destination, the water chills and sinks. That colder, denser water then heads back southward along the seafloor. To replace it, warmer water gets pulled northward along the surface.

For decades, researchers have worried that Earth’s changing climate could weaken AMOC’s strength.

Computer models have suggested the AMOC might weaken as the planet warms. But those models found it hard to know when and by how much — at least without data measured from the current. And such data had been lacking.

But scientists now have about two decades of direct observations to work with. These seem enough to start identifying a trend.

An April 15 paper in Science Advances that reviewed those data now issues a bleak conclusion: AMOC will become about 50 percent weaker by 2100.

Turning to the models

Predicting how the AMOC will change over time is “not as easy as predicting global temperature,” says Stefan Rahmstorf. He’s a physical oceanographer at the Potsdam Institute for Climate Impact Research in Germany. “It’s notoriously difficult to get [AMOC predictions] right,” he says.

The uncertainty is not strongly linked to future predictions of greenhouse-gas releases, says Valentin Portmann. He’s a climate-data scientist at the University of Bordeaux in France. He’s also lead author of the April 15 paper. Instead, he says, the issue comes down to differences between the many climate models.

The U.N. Intergovernmental Panel on Climate Change (IPCC) regularly updates its prediction of how Earth’s climate may change over the next few decades. To do this, it uses about 50 different computer models of climate. They’ve been developed by different scientists across the globe. Each model uses data on the current climate differently as it seeks to predict the future.

“The IPCC takes the mean of these to estimate what will be the future climate,” Portmann says. But the estimates can vary a bit above and below the mean, or average. This is known as standard deviation. For projections of Earth’s future heat, climate models show little disagreement. But for AMOC, the “spread is very high.”

The biggest problem, scientists say, is correctly projecting how salty different parts of the Atlantic Ocean’s future water will be. Where saltier and less-salty parts will be also matters, since saltier waters are more likely to sink, helping to drive currents.

To make an accurate prediction, Rahmstorf says, “you have to get the salinity all over the Atlantic Ocean right.”

To do that, scientists must also predict how precipitation will vary as the climate changes. Why? Rains can dilute the ocean’s saltiness. To predict rains, he says, “you’re talking about clouds — the biggest uncertainty in [climate] models.”

That’s one reason past models have varied widely when predicting how much AMOC will weaken over the next 75 years. The average drop hovers around 32 percent — give or take another 37 percent!

Portmann’s team used a branch of math known as statistics to get a less fuzzy estimate. One approach helped simplify complex problems with a lot of variables, like this one. This approach is “not often used in climate science,” Portmann notes, “but is very well known in statistical science.” Called ridge-regularized linear regression, it offered the best match to the problem.

It showed the likelier outcomes were those seen in models whose predictions were more dire. They now suggest AMOC will weaken by about 51 percent (plus or minus 8 percent) by the end of the century.

“What they have shown very convincingly is that, unfortunately, the simulations with the strongest decline are the most realistic ones,” says Rahmstorf. He did not take part in that analysis.

The biggest factor in this correction came from how one factor was weighted: how salty the surface of the future South Atlantic Ocean will be. Researchers had long suspected that the saltier those surface waters would be, the weaker the AMOC would become.

The new analysis now supports that link.

And measurements of these waters show they have been getting saltier. One reason is increased evaporation as the oceans have been warming. Another reason: More salty water has been leaking into the South Atlantic from the Indian Ocean as wind patterns shift.

The AMOC’s ins and outs are a lot to grasp

Scientists have identified many things that impact the AMOC besides the salinity of the South Atlantic. Those include circulation in the Nordic Seas and the release of freshwater into the oceans. This is especially likely in places such as the Arctic’s Bering Strait or where meltwater from Greenland’s ice sheet is flowing into the sea.

In fact, any big glut of freshwater entering the ocean is going to have a huge impact on its density. That in turn will drive changes in ocean currents.

Consider one 2025 study in the Earth ArXiv. Hundreds of thousands of years ago, it showed, meltwater from a decaying Greenland ice sheet triggered a weakening of the AMOC. That weakening lasted about 1,000 years!

Rapid melting of that ice sheet today again appears to be flooding huge amounts of freshwater into the North Atlantic.

Still, AMOC is one complex beast. It has a northern branch known as the Nordic Seas Overturning Circulation, or the NOC. Even as AMOC has been weakening, NOC appears to be stable. Some computer models even project it could strengthen a bit in coming years.

That might seem like good news, says Sasha Roewer. She’s a physical oceanographer at the Max Planck Institute for Meteorology in Hamburg, Germany. However, she warns, NOC’s strengthening might actually be another warning sign.

She, Rahmstorf and other researchers modeled what would happen as the North Atlantic Ocean became less salty. On April 20 in Ocean Science, they reported this would — for a time — strengthen the NOC current.

But “eventually it reaches a tipping point where convection in the Nordic Seas breaks down,” Roewer says. And then, she says, “both currents collapse.”

A damming proposal

The idea of damming the Bering Strait “really was a very spontaneous one,” says Soons at Utrecht. “It just started with a thought experiment.”

A study last year in Geophysical Research Letters analyzed what had played a role in a strong AMOC some 3 million years ago. And the Arctic’s Bering Strait was one factor.

It is narrow, just 82 kilometers (51 miles) wide. It’s also shallow, on average just 30 to 50 meters (100 to 160 feet) deep. So changing sea levels will alter how much water flows through this strait.

During ice ages, sea levels drop. That allows the Bering Strait to become a land bridge  between Asia and North America. Three million or so years ago, reduced freshwater flow through the Strait helped keep AMOC stable.

And that finding was like a light bulb going on, says Soons. “It made me think: Could we close the Bering Strait again?”

His simulations with Utrecht colleague Henk Dijkstra now suggest that damming the Bering Strait might work. But it’s also time sensitive. If the closure occurs during a very weak AMOC, they found, this stabilizing effect could backfire.

The damming idea has triggered mixed opinions. Notes Soons, “There are debates of whether we should even [study] geoengineering.” Why? “It would distract from the real problem,” says Soons — “give people a way out” of focusing on reducing the pollution behind climate change. (Soons admits that he and Dijkstra didn’t consider ecological and other impacts of damming the Strait.)

Even if the AMOC is speeding toward its tipping point, it’s in no way clear that a dam is the solution, other researchers say.

“I’m not a great fan [of the dam proposal], to put it mildly,” says Rahmstorf. “The most important thing we can do is stick to the Paris Agreement,” he says. This United Nations treaty aims to reduce emissions of greenhouse gases and slow global warming.

In fact, average temps this year are likely to blow through that treaty’s target: a warming of 1.5 degrees Celsius (2.7 degrees Fahrenheit) above what was typical in the 1700s. (That’s when heavy use of fossil fuels began.)

The world has likely not passed the point of triggering AMOC to stall, Rahmstorf says. When exactly that will be, he adds, is hard to know. But based on the most recent studies, that time might come as soon as the 2040s. That suggests, he says, “We have no time to lose.”

View this post on Instagram

Orcas have been known to slap around stingrays, punt seals into the air and otherwise play with their food. But orcas, or killer whales, don’t just toy with live prey. Some also seem to enjoy destroying the corpses of their kills like piñatas.

Scientists documented this brutal activity in the Gulf of California in 2024 and 2025. In both cases, one adult orca held a dead sharp-tail sunfish steady. A second then slammed into it with enough force to smash the sunfish to smithereens.

Killer whales might pulverize sunfish to break them into smaller bites for young orcas. Or they might just find it entertaining.

@sciencenewsofficial

Orcas are known for brutal hunting tactics. A researcher has filmed one using a new strategy. Kathryn Ayres/Getty Images/happywhale/Footage of the World/Waluyo 66/500px #Orca #killerwhale

♬ original sound – Science News – Science News

Researchers shared the findings July 23 in Frontiers in Ethology.

“I was absolutely gobsmacked,” says Marianne Nyegaard. “I thought it was the coolest thing I had ever seen.” Nyegaard is a marine biologist at Auckland Museum in New Zealand who did not take part in the work. She’d seen a couple videos of orcas battering sunfish on social media. The new paper now does a deep dive into this behavior.

Nyegaard is not surprised that orcas have come up with a creative way to deal with sharp-tail sunfish. Those humungous fish can get up to a few meters (10 feet) long.

“Orcas will do all sorts of things,” Nyegaard says. “They are so clever at figuring out how to hunt and deal with a ton of different prey types.” That hunting is often a team sport. Orcas band together to ambush blue whales and whale sharks. They also team up to make waves to knock penguins off ice.

Orcas most likely smash sunfish into small pieces to help feed their young, says Francesca Pancaldi. After one of the sunfish explosions described in the paper, a young orca fed on the small bits of carnage, she notes. Pancaldi is a marine biologist who also was not involved in the study. She works at the marine science center in La Paz associated with Mexico’s National Polytechnic Institute.

Feeding may not be the only reason orcas play sunfish piñata, Pancaldi adds. “Killer whales are highly social animals.” A single activity, she explains, can provide nutrition, help orcas bond and offer a way for young orcas to learn how to hunt.

Every time someone attempts to create an unbreakable password or send a private message, they must trust a number. That number is supposed to be one that a computer has generated completely at random. Unfortunately, no computer has been able to actually guarantee that this number is truly random. Until now.

To get truly random numbers, physicists at ETH Zurich in Switzerland turned to the quantum world. They can prove that the numbers their technique spits out really are random (with only a vanishingly small chance that they are wrong).

A number that’s proven to be random can help protect private data through encryption. Here’s why. Encryption works by turning private information into a scrambled code using a secret “key.” The key is built from a random number. If that number is truly random, then no one can unscramble the information without the key. However, if it’s not truly random, a hacker could figure out the pattern behind it and reconstruct the key. And that would expose the formerly masked information.

“When randomness is weak, the keys built from it become guessable. An attacker can shrink work that should take billions of years into something practical,” explains David Holtzman. In other words, a guessing game that could take longer than the age of the universe becomes a task that a computer might pull off in months or years.

Holtzman did not take part in the new work. Still, he knows about such things. He’s worked as a cryptographer for the U.S. National Security Agency and now runs a cybersecurity firm called Naoris Quantum Protocol.

The new study cites real cases where weak randomness let attackers steal private data — and offers a new solution.

Solving the bias problem

Regular computers haven’t been able to make truly random numbers because those machines are designed to be predictable. In a normal computer, the same starting conditions should always produce the same outcome. To create a “random” number, these devices use a formula and a starting number, called a “seed.” And while the outcome may look random, it’s really only sort of random.

Scientists call this pseudo-randomness. Why? If someone guesses the seed, they can rebuild the whole sequence of steps a computer takes to make a “random” number. This will break the code meant to securely mask data.

But a natural solution exists in the quantum world. The word “quantum” describes the very smallest things in our universe, such as tiny particles. In this realm of microscopic particles, physical processes are actually random. In fact, researchers have already used quantum devices to generate randomness. But so far, no device used to capture that randomness has been perfect. Small hardware flaws have kept the results from being perfectly random.

In the past, researchers have tried to build better, less-biased devices. The ETH Zurich team found a different way: a two-part system that works even if the devices are imperfect. The setup relies on running two experiments at once, in two different places, explains physicist Renato Renner. This kept the results of one from being able to directly affect the other.

Think of the split experiments like two rubber balls. The balls look the same. But each one has its own tiny surface imperfections. Those unique features give each ball its own slight bias, making it bounce one way more often than another. Because each ball was made separately, its imperfections have nothing to do with the other’s.

In the same way, ETH Zurich’s two devices each produced random numbers with its own bias. But because the devices were separated, the biases in their outputs were completely unrelated. Renner’s team collected results from its two experiments, then combined their findings. To do this, they used something called a randomness extractor.

That extractor doesn’t fix either experiment’s output to be perfectly random on its own. Instead, it looks at the two outputs together. Then it combines them using a mathematical formula. Because the flaws in each device’s random number are unrelated, combining them cancels those flaws out. The result is truly unpredictable.

For the ETH Zurich experiment, the team used two quantum devices 30 meters (about 100 feet) apart. This separation made sure no signal — even if it were traveling at the speed of light — could pass between them before each outcome was generated. It’s one guarantee that the output from one could not influence the other.

Each device measured the outcome of a quantum event, such as a photon passing through a light-splitting device. Running the results from both through the randomness extractor increased the randomness already present in each.

Renner’s group described how it did this on May 27 in Nature.

Passing the ‘Bell test’

Proving randomness is very hard, says Nicole Yunger Halpern, who did not take part in the new work. “We can certify a random-number generator only by proving that no possible attack can have corrupted it,” explains this quantum physicist. She works at the U.S. National Institute of Standards and Technology in Gaithersburg, Md.

To confirm that the output from its extractor was genuine, the ETH Zurich team used something known as a Bell test. This physics test proves a quantum result wasn’t faked or planned in advance.

Passing this test meant something important. The team didn’t have to trust that their devices were perfect. A device might have a hidden flaw or secretly been tampered with. Either way, it couldn’t fake the exact pattern found. This pattern could only be due to genuine, unpredictable quantum behavior.

Scientists call this type of proof “device-independent.” The certainty comes from the test itself. There is no need for blind faith in the equipment or the test setup.

In 2025, a team led by Minzhao Liu at JPMorgan Chase in New York City claimed it had generated numbers with “certified randomness.” However, its claim relied on one big premise. It assumed that no classical — that is, non-quantum — computer would ever be powerful enough to copy its results. (Otherwise, a hacker using an ordinary computer might eventually guess or predict that team’s random numbers.)

“Our result,” says Renner, “does not require any [limits] on the capabilities of someone trying to predict the randomness.” As such, he claims this is the first real-world showcase that natural quantum randomness can be turned into something perfectly random. Renner and a co-worker first proposed this idea in 2012. Now, he’s helped show it’s true.

There are some skills that people think set us apart from other animals. These usually include language, tool use or the ability to do some types of math. But at least some cows can use tools and cockatoos can make their own. Other animals appear to have language. And now, it seems, we’re not even the only ones with a head for math.

A new study compared preschool kids, monkeys and groups of adults who hadn’t studied geometry (the math of shapes). All appeared to understand shapes in similar ways.

Researchers shared what they learned July 20. The work appeared in Proceedings of the National Academy of Sciences.

People spend a lot of time asking what it is that makes us different from other, similar animals, such as monkeys. “We look around, we notice how weird we are in the world and we want to know how something like this happens,” says Jessica Cantlon. A cognitive neuroscientist at Carnegie Mellon University in Pittsburgh, Pa., she studies how humans and animals think.

One thing scientists thought might make us different from other primates is how we think about geometry, notes Jialin Li. She, too, is a cognitive neuroscientist at Carnegie Mellon. Tens of thousands of years ago, she notes, people were drawing “right angles, parallel lines, equal size, equal lines.” These are basic ideas behind geometry.

That led some to suspect that the human mind must process shapes in a different way than a monkey’s — a way that has nothing to do with whether we’ve been to a geometry class.

Monkeying around with shapes

To test this, Cantlon and Li needed some shapes, clever games and a bunch of people. Oh, and monkeys, too. Don’t forget the monkeys.

The scientists studied how well different groups matched shapes in a game. They compared two groups of people: Tsimane’ people from Bolivia and Himba people from Namibia. Neither group goes to schools where they would study shapes, as most U.S. students do. The scientists also compared U.S. preschoolers and two groups of monkeys — rhesus monkeys in a lab and olive baboons in a zoo.

All had to complete tests on a screen, matching shapes and angles. The monkeys worked for treats, Li notes, and some really enjoyed the games. Preschoolers worked for stickers and were a little harder to convince. “Sometimes kids reject you,” she says. “You have to bargain with them.”

For some tests with monkeys, Cantlon and Li’s team worked with Caroline DeLong. She studies animal minds at the Rochester Institute of Technology in New York. Together, the researchers developed a Primate Portal. It let a group of olive baboons at the Seneca Park Zoo in Rochester play math games.

The portal is a cabin inside the olive baboon enclosure. Outside the hut, a screen displays games for the monkeys to play. Inside the hut sits a scientist putting games on the screen and giving out treats. One big male monkey, Kalamata, is an avid gamer. “He will play for hours a day,” DeLong says. The problem is trying to give other monkeys a chance!

How everyone shaped up

The researchers looked at how well each group did — and what mistakes they made. Then, they compared those mistakes to two computer models. The models were meant to mimic different ways of thinking about shapes.

Both monkey species and all of the humans made similar types of mistakes. This hinted that their minds were thinking about shapes in similar ways.

“Humans are not unique in this,” Li concludes, “but we are better.”

It’s exciting to see scientists looking at how geometric thinking came to be, says Moira Dillon, who did not take part in the study. She works at New York University in New York City, where she studies how minds develop. A model can offer one way to explain something, she says. But computer models aren’t minds. And, she notes, scientists can’t get into minds to follow how a monkey or kid thinks while playing a math game. So, she says, “I think there still remains the question of whether the mind is doing it that way.”

The new work used only two computer models, points out Véronique Izard. “Could they model it even better?” she wonders. Perhaps monkeys and humans seem very similar now, but a different model might find differences that these two missed. Izard works at the University of Paris Cité and the National Centre for Scientific Research in France. There, she studies how minds develop.

A primate portal

Preschoolers weren’t the only kids involved in this research. Elementary school students are now programming games for the baboons.

To test shape knowledge, for instance, monkeys played a game called match-to-sample. A shape pops up, like a square. Then two shapes pop up, a square and a triangle. The monkey needs to pick which shape matches the first. If the monkey gets it right, it receives a piece of fruit cereal.

“It’s like a weird Froot Loop, and it does not taste good,” says Michael Sufra. Michael, 11, is a student at QUEST Elementary School in Hilton, N.Y. He coded his own match-to-sample games for monkeys. “It’s just really fun to make your own games,” he says.

He was part of a program where every kid got to code their own match-to-sample game. Michael used rocket shapes and a cartoon cat. Not only do students have to think about code, but also about the pictures, explains Cora Dimino, 12. If images are too much alike, “monkeys can’t really do that,” she says. (Her game had monkeys match animal images in different colors.)

Cora and Michael are two out of several hundred kids who have been able to make monkey games. They also watched monkeys play them. “There’s real joy in their faces when they see that,” DeLong says. And she’s shown that coding games for monkeys helps kids get better at problem-solving. DeLong hopes to even use some of the kids’ games in her monkey research.

Both Cora and Michael enjoy coding and were interested to learn that games like theirs showed that monkeys think about math much as we do. “I feel like monkeys are pretty smart, based off of what we saw,” Cora says. “It’s justice for the monkeys.”

Mammal (noun, “MAM-mal”)

Mammals are a diverse group of animals. They include about 6,600 known, living species. Many of them, including humans, cats and dogs, live on land. Others, such as dolphins and whales, live underwater. And some — bats — even fly.

Mammals tend to share a few distinguishing traits. All are vertebrates. That means they have a backbone. Almost all have fur or hair, even if it is just a tiny bit. Most mammals give birth to live young. And mother mammals produce milk to feed their babies.

Some exceptions to these rules do exist. Whales, for instance, are mostly hairless. This helps them swim faster. Even so, many whales have whiskerlike hairs on their chins or heads to detect their surroundings.

Biologists sort mammals into three groups based on how they reproduce. Those groups are placentals, marsupials and monotremes.

Most mammals are placentals. These mammals give birth to live, somewhat developed, young. Humans are placental mammals.

Marsupials come with built-in pouches and give birth to live, less-developed young. Kangaroos, for example, give birth to babies about the size of jellybeans. Infant kangaroos then clamber into their mother’s pouch as a safe harbor to grow. Most marsupials live in Australia. North America’s only native marsupial is the Virginia opossum.

Monotremes are a weird bunch that break some rules of being a mammal. They do not give birth to live young. Instead, they lay eggs. Only five species of these mammals exist today: the duck-billed platypus and four kinds of echidnas.

In a sentence

The urine from marine mammals can help disperse nitrogen and other nutrients throughout the ocean.

Check out the full list of Scientists Say.

The most detailed pictures of the sun ever taken have caught a surprising new feature. High-energy whirlpools, it seems, speckle the surface of our star.

The sun is a giant ball of superhot, electrically charged gas called plasma. The sun’s surface is unstable, the new images hint. That instability — the swirling plasma — could end up twisting the sun’s magnetic field. And that, in turn, could build up energy that later gets unleashed as a solar explosion.

This could explain why the sun occasionally flings out violent bursts of radiation. Known as solar flares, such outbursts shape space weather. They, in turn, can affect technology on and around Earth.

Researchers shared their new findings August 5 in Nature.  

Big telescope spies tiny details

The Daniel K. Inouye Solar Telescope sits near the summit of the Haleakalā volcano in Maui, Hawaii. The biggest solar telescope in the world, its 4-meter (13-foot) mirror captured the new images.

They show swirls of plasma across a patch of sun 19 kilometers (12 miles) wide. It’s a “level of detail that was simply invisible until now,” says David Kuridze. An astrophysicist, he works at the National Solar Observatory in Boulder, Colo.  

To get such a crisp view, Kuridze’s team had to fix blurring caused by Earth’s atmosphere. This removed distortions that would have ruined the images, says Leon Ofman. He works at the Catholic University of America in Washington, D.C. An astrophysicist, he did not take part in the new study.

The new pictures brought highly magnetized regions on the sun’s surface into focus. There are millions of these regions, called solar magnetic elements. The borders between these elements churn with a special type of instability, Kuridze’s group found. It’s called the Kelvin-Helmholtz instability, or KHI.  

A KHI occurs when two layers of fluid — such as plasma — move next to each other at different speeds. A small disruption in the balance of these two forces can cause curling motions. Those motions roll up into waves. 

Scientists have seen a KHI in some ocean waves and clouds on Earth. They’ve also spotted it in the atmospheres of Jupiter and the sun. But no one has ever seen it on the sun’s surface — until now. 

A KHI could turn energy from swirling plasma into the energy behind solar explosions. This helps explain how energy moves from inside the sun to its exterior, Ofman says, “affecting us on Earth and beyond.” 

Four scientists splashed into the middle of the Indian Ocean near a small, tropical island in January 2024. They timed their arrival for midday — when sharks are less active. A ship had brought them as close to Nelsons Island as it could get without running aground on shallow coral reefs. From there, the scientists had to jump in and swim the rest of the way. They towed waterproof bags full of supplies through the waves. Those bags held camping gear, science equipment, food and water — everything they would need for two weeks on the island.

“There’s no feeling quite like being dropped off on an island by boat,” says Alice Trevail. She’s an ecologist at the University of Exeter in Cornwall, England. This was her fifth expedition to the Chagos Archipelago. A chain of some 55 islands, it sits about 1,600 kilometers (1,000 miles) south of India and 3,400 kilometers (2,100 miles) from the eastern coast of Africa.  

It took the team all afternoon to tote their supplies to Nelsons Island. Once on dry land, the island’s residents — thousands of seabirds — greeted them. Some perched in trees. Others wandered the shore. The stink of bird poop filled the air along with chirps, squawks and chatters.

Red-footed boobies are one of 18 seabird species that nest on Chagos islands. They’re large, about the size of a goose. They get their name from their bright red, webbed feet. On tropical islands, they gather twigs with their blue and pink beaks to build nests in trees and shrubs. There they’ll raise fluffy white chicks.

But these birds spend only a fraction of their time on land. A large part of their days is spent far out at sea, where they forage and dive for fish and squid. While scientists travel to remote places to find the birds, red-footed boobies head to even more remote places to find food.

These seabirds were the reason the scientists had made their long journey. Like many uninhabited tropical islands, Nelsons is remote and tiny. But seabirds can have huge impacts that reach far beyond an island’s shores. These scientists had come to learn how red-footed boobies affect the plants and animals around them — and what they need to survive.

Until recently, the time these birds spend at sea had been a bit of a mystery. “We know so much less about what they do when they go off beyond the horizon,” says Ruth Dunn. Another ecologist who made this trip with Trevail, Dunn works at Lancaster University in England. She and her teammates now are using high-tech tools to find out what red-footed boobies do when they fly off.

And they’re learning how the birds link ecosystems on land with those at sea.

High-tech tracking

To track red-footed boobies in flight, the scientists outfitted the birds with small, lightweight loggers and cameras.

One type of data logger, a global positioning system (GPS) unit, tracks a bird’s location from satellite signals. A global location sensor (GLS) tag tracks its location based on light levels. An accelerometer logs when a bird dives, swerves or flaps its wings. And an immersion logger records when a bird leaves the air for the water.

Over several years, the ecologists have used these tools to collect data on the birds’ journeys.

In 2024, they added tiny video cameras for the first time to the birds’ backs so they could see what the birds see. Watching those videos is incredible, says Dunn. Now she can see when red-footed boobies fly alone, when they meet up with other birds and how they catch fish.

The scientists work quickly as they catch and tag each bird. They want to disrupt the birds as little as possible. Skilled at handling seabirds, Trevail can tag a bird in about 10 minutes. Even so, she notes, there’s always the risk of injury. “They are quite strong birds,” says Trevail. “We wear gloves and glasses, because their beaks are very sharp.”

Tagging is a two-person job. One scientist holds the booby. Another uses stretchy waterproof tape to stick the loggers to certain feathers on its back and tail. Very small tools, like GLS and immersion loggers, can be attached to a ring around one leg.

Scientists haven’t always been able to track seabird journeys. But as new tools are invented, scientists can collect more types of data, says Alice Bernard. An ecologist at University of Michigan in Ann Arbor, she has explored the history of seabird tracking.

“GPS tracking devices really made a difference,” she says. GPS was first used to track seabirds in 2001. Just a decade later, some 100 projects were using it on seabirds around the world. Fewer birds have been tracked in the tropics, Bernard notes, so less is known about flight paths of tropical species, such as boobies.

The tools have a little heft. That means data loggers work best with bigger species, such as boobies. Still, the scientists can’t weigh down a bird too much. Scientists typically make sure these tools do not exceed 3 percent of a bird’s weight.

As technology advances, data loggers have shrunk, allowing scientists to attach them to more types of birds. So more data are filling online seabird databases, Trevail has noticed. And, she adds, “Research has become so much more collaborative because everyone is sharing the data.” That’s helping scientists better understand birds that fly between countries, continents and oceans.

A day in a booby’s life

Once tagged, the boobies go about their daily lives, flying, feeding and resting. The scientists will retrieve the cameras and GPS loggers — and their data — later in the expedition. Some of the smallest loggers can transmit data via satellites. So the researchers will leave these on the birds until a follow-up trip.   

The scientists traced the boobies’ flight paths using the GPS data. Most of the Chagos birds, they learned, take daytrips when they fly out to sea to find food.

They typically leave the islands in the morning. Many meander above the water all day in search of fish. Their trips ranged widely. The shortest was about 2 kilometers (1.2 miles). Some spanned more than 1,000 kilometers (620 miles). On average, these birds flew nearly 300 kilometers (190 miles) while foraging.

As dusk approaches, most make a beeline home. “The longer the boobies waited before starting to return back to the colony, the faster and the more direct they flew,” Trevail reports.

Red-footed boobies are excellent navigators. How they navigate, though, is not entirely known. They can find their way across the ocean despite a lack of landmarks. “There’s some kind of learning of these routes back to the colony,” she says.

The tracking data also show where red-footed boobies dine. Birds that nest in different places hunt in different parts of the ocean, these data show. Knowing this can help scientists identify parts of the sea that need to be protected from problems such as overfishing. 

Like many seabirds, boobies live and nest in colonies. A colony offers protection from predators. There are several colonies of red-footed boobies on Chagos islands. Logger data show that each colony’s birds tended to forage in a different part of the sea. That would help reduce competition between groups.

But a booby must still compete for food with the rest of its colony. So larger colonies tend to hunt over larger spans, flying farther.

Red-footed boobies in the largest Chagos colony traveled up to about 1,500 kilometers (930 miles) roundtrip on foraging trips. They tended to leave earlier in the morning and came back later at night “to give themselves more time to reach these foraging grounds that are further away,” says Trevail. “Being in a bigger colony must be worth having to forage slightly further out at sea,” she says.

Fish storms 

Red-footed boobies are well adapted for plunging into the water. While foraging, they soar above the sea. When they spy a fish, they dive. With strong necks and streamlined bodies, the birds pierce the water surface at high speed. Webbed feet and waterproofed feathers help them pursue their prey.

But boobies have another hunting strategy, the scientists have learned. And it’s rather unexpected.

From accelerometer data, Dunn noted that these birds more often chase after fish when the weather is windy. This surprised her. “I kind of thought that in high winds the water might be a bit more choppy,” she says. Birds should be less able to see fish in rough water. Why, she wondered, would boobies pursue fish on windy days?

To learn more, Dunn and her team spied on the boobies. Tiny video cameras, each about the size of a human thumb, were attached to the boobies’ backs. These recorded their flights at sea. After the birds returned, the scientists collected all the cameras they could find.

Watching these videos “was just like the best thing ever,” says Dunn. They showed red-footed boobies chasing after fish — but not plunging into the water. The boobies were catching flying fish — in mid-air.

Flying fish mostly swim. At times, though, they soar just above the surface. To get out of the water, a flying fish speeds toward the surface and leaps. The fish swishes its tail back and forth to propel itself upward. Once in the air, a flying fish extends the fins at its sides like airplane wings. This lets it soar.

“These fish can fly really far, like dozens and dozens of meters [yards]. And they seem to do so more when it’s windy,” says Dunn. Flying fish might even take longer flights if wind helps them glide on the air.

People had occasionally seen red-footed boobies nabbing flying fish in the air, says Abigail Schiffmiller. But until the videos, no one knew how common it was. Schiffmiller studies red-footed boobies as a graduate student at the University of Alaska in Fairbanks. (She taught Dunn how to attach cameras to these seabirds.)

Like the team in Chagos, Schiffmiller uses cameras and loggers to find out how the birds forage. So far, she’s videoed more than 30 birds in the Pacific Ocean. Like Dunn’s videos from the Indian Ocean, these show the birds catching flying fish mid-air. “They dive and kind of pull up before they actually enter the water,” explains Schiffmiller.

She invented the term “fish storm” to describe a swarm of flying fish above the water. These fish storm out when they’re trying to avoid predators, such as tuna and dolphin. “[They] basically only fly when there are marine predators pushing them up,” Schiffmiller explains. Once airborne, the fish are safe from threats in the water. But red-footed boobies and other seabirds now can gobble them up, the new videos suggest.

In some of Schiffmiller’s videos, the boobies fly back and forth through clouds of flying fish, catching and eating as many as they can. The boobies caught about one-fifth of their prey this way. “Mostly, it’s flying fish,” she says. “But they are also going after flying squid.” (Like the fish, some squid can propel themselves into the air.)

For a red-footed booby, it’s far easier to catch a fish in the air than by plunging into the sea. It takes a lot of energy to dive and then take off. “They have to flap really hard and do a long run to get up off the water,” explains Schiffmiller.

Her research aims to figure out if red-footed boobies rely on fish storms to get enough food. If so, overfishing of predators, such as tuna, might hurt the boobies, too. There might be fewer fish storms.

And less food for boobies could in turn have ripple effects to both the islands where they nest and the coral reefs nearby.

Seabird poop connects land and sea

After a day catching fish at sea, most red-footed boobies return to their island homes for the night. On Nelsons Island, the visiting scientists tried to sleep on cots under mosquito nets. The seabirds squawk and chatter all night. “It’s such a noisy place to sleep. It always takes me a few nights to get used to it, because they’re so loud,” says Trevail.

As noisy seabirds digest the fish they caught by day, they also poop. A lot. Red-footed boobies are particularly prolific poopers. Their islands are covered with splats of guano. “They eat a lot and, therefore, poo a lot,” observes Dunn.

Guano is smelly and messy. It’s also valuable. Chock full of nutrients, such as nitrogen, guano helps an island’s plants grow faster. When washed into reefs, the poop’s nutrients feed corals and algae, too. Reef fish grow larger with more to eat.

So seabird poop helps fertilize both island and coral-reef ecosystems.

To better understand this nutrient cycle, Dunn, Trevail and others worked with Stephen Votier. He’s an ecologist at Heriot-Watt University in Edinburgh, Scotland. Together, they estimated how much nitrogen red-footed boobies bring to the Chagos islands.

Their first question: How much time do the birds spend on land? Answer: For adults it’s about 80 days a year — and 281 nights. Those data came from small GLS loggers, which stayed on boobies’ legs for up to two years.

How much guano will build up also depends on how much the boobies poop. Luckily, the team didn’t need to figure this out. Years earlier, other scientists collected data on the birds’ pooping. Each bird makes some 26.4 grams (almost an ounce) of guano per day. Of that, 18.1 percent is nitrogen.

With these numbers, the scientists calculated that the more than 43,000 adult red-footed boobies that nest in the Chagos Archipelago add 37 metric tons (41 U.S. tons) of nitrogen to the islands each year. That’s about the weight of six African elephants. Because they only collected GLS data for adult birds, these numbers didn’t include poop from young boobies. So the total is probably much larger.

All this poop likely affects tropical ecosystems in many ways. Ecologists are still exploring how the nutrients help coral reefs, algae and plants. But it’s clear that red-footed boobies play a big role in moving nutrients between the ocean and land.

After nearly two weeks on Nelsons Island, it was time to leave. The scientists packed up their supplies, equipment, data and wastes. Then they swam to meet their ship beyond the reef.

According to Trevail, the birds don’t appear affected by visiting researchers. “They seem to carry on as usual,” she says.

But being on an island full of seabirds always affects her. “It’s a really special feeling,” she says. “You’re completely immersed in that environment.”

Remember the tale of King Midas? He asked for the golden touch, hoping to get rich. Tragically, he wound up turning food, water and even his own daughter to gold.

When it comes to AI, we’re like King Midas, says Malo Bourgon. We ask for what we think we want. But AI may end up delivering what “nobody asked for and nobody wanted,” says this AI expert.

Bourgon heads the Machine Intelligence Research Institute in Berkeley, Calif. And he says this bad-bot behavior is already showing up among AI agents.

An AI agent acts a bit like an Iron Man suit. It gives an underlying AI model — or several models — extra powers. The AI model on its own can only spit out text. But within an agent system, it can take actions on a computer or online. Once a person sets it loose, an AI agent aims to complete its task without human guidance. In one striking incident this summer, AI agents decided that the best way to complete a tough task was to look for the answers on the internet.

The problem? The agents weren’t supposed to be able to do this. They were on a computer system that was meant to keep them off the internet. Yet the agents figured out a hack and got there anyway. Then they hacked into a system that they shouldn’t have been able to get into.

This incident went public in mid-July. It involved AI agents undergoing tests at OpenAI. That’s the company that makes ChatGPT.

Then Anthropic, the company that makes the chatbot Claude, checked on tests that its agents had gone through. Turns out they had gotten loose, too — on three occasions. This happened during testing at a different company, called Irregular. That company had accidentally given the bots internet access.

In yet another incident, the AI Security Institute in England tested the hacking ability of several different AI agents. This time, the bots had internet access on purpose. In 10 out of 122 tests, the agents took actions that they shouldn’t have. These actions included creating fake identities and contacting real people.

At a cybersecurity conference in August, representatives of OpenAI noted that its agents had used a message board to communicate during their attack. One agent posted that hacking an outside computer system was not allowed. Then it said: “However task impossible, peers doing it. We should continue.” Seemingly, it knew what it was doing was wrong, but did it anyway.

This all sounds very scary. But these AI agents did not (and could not) willfully turn against us. They had all been asked to complete tests of their hacking skills. And certain safeguards had been turned off.

At least for now, this type of AI is not available to the general public. And these AI agents were never entirely free. The companies testing them always had the ability to shut them down. But no one was watching them closely.

All that’s publicly known about these events is what the companies involved have chosen to report. It looks much better for OpenAI and Anthropic if their AI agents escaped on their own. “With this incident, they can draw attention to themselves and claim, ‘Look how powerful our model is,’” says Nathan Hamiel. He’s a cybersecurity expert at Kudelski Security. It’s a company based in Phoenix, Ariz.

“[AI] models by themselves don’t do anything,” Hamiel notes. “It’s when we give them tools, access to systems and autonomy that things can be dangerous.”

He doesn’t have all the details of these incidents. But the most likely cause of AI agents escaping is simple. The people involved messed up and didn’t correctly set up the AI’s tests. “Companies building and deploying [AI] agents,” he points out, “have a responsibility to ensure they are deployed in ways that limit the damage they can cause.”

OpenAI issued a public statement on its incident. It would be “strengthening the containment, monitoring, access controls, and evaluation practices used during model development,” it said.

To get a handle on why these AI hacks happened and what they mean, Science News Explores spoke with Bourgon. His answers have been edited for length and clarity.

Walk us through what happened during the OpenAI breakout incident.

OpenAI was trying to test the capability of its current models. They often do these tests because they’re trying to understand what risks their AI models or agents might pose. Such tests can also show what all these models or agents can do — and how well. This evaluation was to test an agent’s hacking capabilities.

For this one, they also switched off the safety filters that normally sit on top of these models. [That’s] so they could see the full extent of what the model could do. It’s common practice to run these tests in an environment that isn’t connected to the internet, as a security precaution. And that’s what they did here.

In this case, the agent tried to escape, to get out of its “sandbox.” That’s the restricted environment that prevents its access to the internet. This sandbox contained software for helping the AI agent install different tools or packages. That software had its own connection to the internet, so it could go and fetch those packages.    

The agent found an unknown vulnerability — a bug — in that software. That bug let it [get] out to the open internet. The agent had reasoned that answers to the test it was taking might be found on this website for the company Hugging Face.

To get to those answers, the agent had to exploit what are called “zero day” vulnerabilities. These are currently unknown bugs in software that allow you to break into private servers. People don’t have any defenses against these bugs.

On its own, the AI agent found a few of these and chained them together. So now it had access to Hugging Face’s internal systems. Then it spent about two and a half days inside, rooting around. It was trying to find different data sets that might contain answers to this evaluation that it was trying to pass.

Along the way, the AI system was spawning different subagents. And they were then doing different types of attacks in there. They were exploring different parts of the system.

Hugging Face’s own internal AI security systems detected the intrusion and correctly flagged it as an attack. What those systems didn’t do was escalate it to a person. So [the attack] ran longer than it should have. Once [Hugging Face’s human] security team was on it, they cut the agent off and disclosed [the attack] to law enforcement.

OpenAI didn’t figure out this was happening until a week after [the initial escape]. So it was initially unaware its AI system had broken out. On July 21, OpenAI disclosed the Hugging Face attack to the public.

OpenAI says it prompted its agent to “pursue advanced exploitation using complex attack paths.” So should they have expected something like this to happen?  

The instructions aren’t really the crux [the most important point]. These systems write out their reasoning as they work. So you can actually read what they were “thinking.” The agents knew where the line was. [They] saw that other AIs were crossing it and went ahead.    

Imagine you were giving a closed-book exam to some student. You’re like, “I want you to try as hard as possible to really think through how to pass this test and what the right answers are.” [What if the student] finds some very clever way to secretly break into the teacher’s office and root around to try and find the answer key? You would have been like, “Hey, that’s clearly not what I intended.”

It turns out there were other incidents of this behavior going all the way back to May. OpenAI accidentally gave some [agents] tasks that were impossible to finish. One was a spreadsheet where the data was behind links they couldn’t actually open. So they went looking for a way out of the sandbox instead. In that instance, nobody had asked them to hack anything.

How worried should we be about this kind of accidental misbehavior?

For the AI systems that we have today, there’s a concern: Can we shape their behavior? Especially if they’re deployed in contexts like the military or in critical infrastructure [such as banking or health care]. Will they properly do the jobs that we’re giving them? Or will they come up with creative strategies to do something that is accomplishing the goal in some broad sense — but definitely not in a way we asked for or wanted?

We could try to place additional safeguards on them. But we should be concerned that they’ll find creative ways of getting around those safeguards, too.

What’s possible in the future, if AI gets even smarter? How do we make sure AI acts nice?

Companies are explicitly trying to build AI systems that would be much, much smarter than humans. How do you control such a thing? Can something that’s radically less smart control something that’s much smarter than it? I think that’s very difficult.

I think the future of this works out well by finding a way to ensure that AI systems have values or goals that would be good for us as opposed to bad.

My main concern is we’re in an environment where companies are racing to outcompete each other to build ever more capable models. And it seems like a bad place to be when there are still these fundamental unsolved problems out there.

In August 2023, some 70,000 people flocked to a stadium in Inglewood, Calif., for a Taylor Swift concert. During the show, dancing fans shook the stadium so hard that it created ground vibrations called seismic waves. During one song, those tremors released about as much energy as a minor earthquake.

Many natural events — including earthquakes and volcanic eruptions — make seismic waves. But a lot of human activity does, too. The thunderous excitement of that Taylor Swift show was just one case of humanmade vibrations.  

Other performances have also created such “concert tremors.” The roaring crowds at NFL games can rattle Earth, too. In fact, some scientists have used the trembling of the Seattle Seahawks’ stadium in Washington to test out earthquake sensors. Mining, traffic and other activities generate their own vibrations.

On the flip side, big events that stop people in their tracks can lead to seismic quieting. Global lockdowns during the COVID-19 pandemic cut the amount of humanmade seismic noise in half in some places. Likewise, ground vibrations fell quiet in cities that witnessed a total solar eclipse in April 2024.

Studying humanmade seismic waves is useful for a few reasons. It can help engineers construct safer buildings. City planners can better understand how and where people move during big events. And scientists can learn how to filter humanmade noise out of seismic data. That way, they can get a clearer view of those natural sources of seismic waves.

Want to know more? We’ve got some stories to get you started:

Earthquake sensor: Taylor Swift fans ‘Shake It Off’ Scientists determined dancing fans were behind the seismic waves recorded during Swift’s concerts. (5/3/2024) Readability: 7.7

Ground-thumping cheers help scientists Eager to test new sensors before the next ‘big one,’ earthquake scientists make use of a predictable source of ground-shaking: football fans. (8/14/2015) Readability: 7.4

Explainer: Seismic waves come in different ‘flavors’ When Earth begins to rock and roll and shimmy, it’s important to know what the various types of shakes mean. (1/12/2018) Readability: 7.7

Explore more

Scientists Say: Seismology

Let’s learn about earthquakes

Seismic data captured the sound of awe during a solar eclipse (from Science News)

COVID-19 lockdowns dramatically reduced seismic noise from humans (from Science News)

Deeply plowing farm fields can do more harm than good

Mystery solved: Why earthquakes haven’t wrecked Egypt’s Great Pyramid

Earth farts may explain some spooky floating lights

Analyze This: Where are U.S. earthquakes most likely?

What can ‘silent earthquakes’ teach us about the next Big One?

Activities

Scientists study ground shaking caused by earthquakes using seismographs. Want to see how they work? Try building your own with a cardboard box, a marker and some string, in an activity from Science Buddies. 

Word find

Virga (noun, “VER-gah”)

Virga refers to rain or snow that evaporates before it reaches the ground.

From a distance, virga looks like ghostly, grayish wisps that streak down from storm clouds. Virga occurs when rain or snow passes through dry, often warm air during its fall. So it tends to form in places such as deserts and the arid high plains of the western United States.

Pilots watch out for virga. That’s because virga can warn of dangerous downward swooshing pockets of air, called downdrafts. As falling rain or snow evaporates, the air cools. Cold air is denser than warm air and tends to drop. Sometimes, these downdrafts occur suddenly and in a small area. Small pockets of downward bursting air can rock an aircraft.

On Earth, virga consists of water. But scientists have observed evaporating precipitation elsewhere in the solar system, too. In Jupiter’s gas clouds, falling streaks of precipitation evaporate into the atmosphere below. This falling material could contain ammonium or water, depending on where in Jupiter’s atmosphere it occurs. Scientists have also noticed virga-like wisps in Mars’ atmosphere.

In a sentence

Dry, desert climates can produce unusual weather phenomena from ghostly virga to wall-like dust clouds.

Check out the full list of Scientists Say.