However, in its current form, the same AI technologies that add watermarks can also strip them away, other experts warned.

New ways to hide watermarks

The unveiling of the AI program AlphaFold in 2020 was a milestone in protein engineering. The tool, which won Demis Hassabis, John Jumper and David Baker the Nobel Prize in Chemistry in 2024, can reveal the 3D structure of a protein from the sequence of its building blocks, called amino acids. Proteins fold like origami, adopting 3D shapes that dictate their function.

However, AlphaFold's technology also opened the door to generating brand-new proteins that could be used to create biothreats, including viral proteins that can evade our immune systems and existing vaccines.

"Anything that can be used for good can be used for bad," said George Rose, professor emeritus of biophysics at Johns Hopkins University.

Current screening tools identify novel dangerous or restricted molecules by seeing how similar their amino acid sequences are to those of recognized threats. But the problem is that AI could come up with proteins that don't look like anything we've ever seen. That's what the new watermarking technique aims to solve.

The DeepMind team has framed the watermarking effort as a "family" of related techniques they call SynthIDBio. The first member of this family is SynthIDBio-sequence, which encodes watermarks into the amino acid sequences of proteins created by a program called ProteinMPNN.

When users tell ProteinMPNN the desired 3D shape for their protein, it chooses amino acid options that ultimately fold into that shape. SynthIDBio-sequence subtly influences how the program chooses each amino acid according to a secret watermarking "key." Once the sequence is generated, anyone who knows the hidden code can assess how likely the sequence is to have been created under the key's bias. In their recent study, the DeepMind team showed that proteins built with this watermark could still bind to their targets as well as non-watermarked proteins.

A sister technique, called SynthIDBio-structure, instead embeds a watermark directly into the protein's structure. It does this by altering the function of AlphaFold 3, DeepMind's flagship protein prediction algorithm.

When AlphaFold predicts a structure, SynthIDBio-structure gives it a side mission: The AI must make minuscule tweaks to how the atoms in the protein are arranged. These tweaks can be picked up by a detector later, and they're small enough that they don't look like obvious distortion.

In a test, these two systems created hidden watermarks that could be detected with incredible accuracy, flagging more than 99% of watermarked proteins and correctly ignoring 99.9% of non-watermarked proteins.

How to bypass a protein watermark

Live Science spoke to several experts who agreed that SynthIDBio was an impressive technical feat. However, they said it wasn't ready for widespread implementation yet.

The two techniques are "effective but brittle," Bruce Wittmann, a principal applied scientist at Microsoft, told Live Science in an email. That's because, as the DeepMind team also notes in their paper, the watermarks can be easily removed.

"For sequences, watermarks can be removed by 'resequencing' — effectively, altering the protein amino acid sequence such that the watermark is no longer detectable," Wittmann said. "For structures, watermarks can be removed by adding noise to or otherwise modifying the spatial coordinates of the structure." Screening tools wouldn't be able to detect if a watermark had been removed from a protein.

Experts are also skeptical that watermarking would provide a long-term solution for biosecurity.

An indelible watermark could identify who originally made a nefarious protein, or identify when a user was accessing a protein generation software without permission. This might deter some criminals, but terrorists looking to create a novel protein for a biological weapon would be less likely to care about being identified after a successful attack, said Max Brackmann, a chemical, biological, radiological, and nuclear arms expert at the Spiez Laboratory in Switzerland.

Additionally, the vast majority of software for protein prediction and generation is open-source, meaning much of the code is freely available to use and modify. That makes it far harder to bake watermarking into these programs. Wittmann warned that the field would have to shift toward closed-source models before watermarks could become an industry standard.

He added that watermarking could act as one layer of a biosecurity defense shield, rather than offering standalone protection. Top-down regulation would be challenging to enforce. "I imagine it would be very difficult to ensure the regulations were being followed," he said.

Rose was even blunter about regulating the use of generative proteins.

"I'm convinced it won't work," he said. "Once there's a sweet technology that does something, government regulation to keep it under wraps has never been really successful."

Help us improve Live Science Pro: We're always trying to make our content better. Leave us feedback about Pro here.

'> Scientists can now 'watermark' AI-generated proteins to deter bioweapon creation. But will it really work? briefly feared that he and his crew would never return home from their trip around the moon after overshooting their turn-back point.

Commander Reid Wiseman and his three crewmates were supposed to reach 219,667 nautical miles (252,788 standard miles, or 406,823 kilometers) — farther than any humans have ever traveled from their home planet — before turning back toward Earth on the sixth day of their 10-day mission to the moon.

But the odometer on the Integrity spacecraft continued to tick upward as they shot past their intended point of return, leading the astronaut to fear that something had gone terribly wrong, he told an audience here at the Starmus VIII festival earlier this week. (This semi-annual event is held to celebrate achievements in science, exploration, and creativity.)

"Mission control in Houston had told us our maximum distance would be 219,667 nautical miles and that was the furthest we were supposed to be from Earth," Wiseman said. "I was watching this number, and I saw it hit 219,667, and I thought, 'Great, this is the farthest we will be from Earth, and the next number I will see will be 219,666 ‪—‬ we'll be going home.' But about three minutes later, I didn't see 219,666; I saw 219,668. It went up. I was like. 'Huh? That's weird.'

"And then about three minutes later, it goes to 219,669, and that was the moment that I felt real fear because I thought we had just gone too far ‪—‬ we're never going to come home ‪—‬ and so I was very scared," he added.

Then, at last, he saw the counter start to drop. Meanwhile, NASA astronauts Victor Glover and Christina Koch were at the window, doing science observations of the moon, and Canadian Space Agency astronaut Jeremy Hansen was sending reports back to mission control ‪—‬ so Wiseman's moment of terror went unnoticed.

When Wiseman saw the odometer click back to 219,668 nautical miles, he said, "We're going home," and the crew let out a cheer. "I knew there was a good chance we might make it," he added.

In an interview with Live Science's sister publication The Week Junior Science+Nature, Wiseman paid tribute to NASA's Earth-based teams for the ultimate success of the mission.

"Getting home safely is a testament to the engineering rigor and the flight control teams back here on planet Earth, who put this rocket together and then who controlled it as we went out to the moon and came back. Those are the true heroes of this mission. They have real courage."

The Artemis II mission launched from NASA's Kennedy Space Center in Florida on April 1. It was the first crewed flight of the Orion space capsule and the first crewed mission to travel to the moon since Apollo 17 in 1972, taking nearly 10 days to fly around the lunar far side and return to Earth.

Nearly every moment of the crew's activities in the Orion capsule was livestreamed to the world, garnering massive public interest in the mission. The Artemis II crewmembers were awarded the Stephen Hawking Medal for Science Communication at the Starmus VIII festival.

Are you a NASA nerd? See how well you score on our Artemis quiz to find out!

'> 'We're never going to come home': Artemis II commander recalls the most terrifying moment of his 10-day trip around the moon

To understand how crows make decisions using cues in their environments, the researchers trained two carrion crows to complete a visual task on a touchscreen. The birds watched a cloud of white dots moving against a black background. Some dots moved to the left or right while others moved randomly. The crows had to determine the predominant direction of the dots' movement and peck the corresponding target on the screen to earn a mealworm reward.

By changing the proportion of dots moving together, the researchers could make the decision easier or harder for the crow. When most dots moved in the same direction, the answer was fairly obvious for the birds. But when only a small percentage moved together, the birds had weaker visual evidence to base their decisions on.

The crows reached nearly 100% accuracy when many dots were involved and the motion was clear. But even when just 3% of the dots moved in concert, both birds performed significantly better than just guessing, getting the answer right 50% of the time.

The birds maintained high accuracy in different rounds of the experiment, suggesting they were not sacrificing judgement for quickness.

The study did not directly compare crows to humans and monkeys performing the exact same task. However, previous studies suggest that humans and monkeys take roughly twice as long on comparable, self-paced visual motion tasks.

What happens inside the crow brain?

To understand how the birds reached their decisions so quickly, the researchers recorded activity from 248 neurons in the nidopallium caudolaterale (NCL) — a region of the bird brain that's considered functionally comparable to the prefrontal cortex in humans and primates. Birds lack the layered cerebral cortex found in mammalian brains, but the NCL performs similarly advanced functions.

In primates, neurons accumulate sensory information and their activity increases until it hits a critical threshold and the brain reaches a decision. The new findings suggest that crows use a comparable process, the researchers concluded.

"The bigger implication is that a brain does not need a mammalian neocortex to perform sophisticated computations," Nieder said.

The team could use neural activity to predict the crows' eventual choice "as early as 70 milliseconds after" the dots appeared, Nieder said. That was roughly 170 milliseconds before the crows' behavioral responses, meaning their pecks at the screen.

The speediness of the crows' decision-making may reflect the demands of the birds' daily lives.

"Flight places enormous demands on visual processing and sensorimotor [sensation and movement] coordination," Nieder said. "Birds must navigate three-dimensional environments, avoid obstacles, and react quickly to moving objects. Finding food and capturing prey can also require rapid, precise decisions. These pressures may have favoured fast information processing."

But for now, scientists don't know why crows seem to outpace primates in this particular type of decision-making. The study included only two birds and no direct comparisons to primates, so more work could be done to confirm the results and explore that question further.

For Nieder, observing the faster decision-making process in crows raises a broader question about how scientists define intelligence.

"Rather than assuming that the mammalian brain is the blueprint for intelligent behaviour," he said, "we can study different species to discover which computational principles are shared across animals and which reflect adaptations to their particular environments."

Are you a bird nerd? Find out with our bird quiz!

'> Crows may make lightning-fast decisions, sometimes twice as fast as humans, study hints

The brain's shared map brings different number formats into a common system based on the quantities they represent, the researchers concluded.

"For me, it is how the brain gets from a symbol to its meaning," said Abiot Derbie, a cognitive neuroscientist and founder of the company LifeByLogic, which makes psychology assessments.

The fraction "¼" contains the digit "4," but the fraction's value is much closer to zero, he noted. "The activity patterns reflected that numerical relationship, going beyond the digits themselves," Derbie, who was not involved in the research, told Live Science in an email.

The idea that the brain represents numerical magnitude in the parietal lobe is not novel. However, scientists mostly understood how the brain dealt with whole numbers, while its strategy for fractions was not well understood. The new work builds on that existing research by exploring how both numerical magnitude and number type are processed together.

"To the best of our knowledge, this is the first study to investigate the geometry of neural representations in a task involving both integers and fractions," the researchers wrote in their report. "In the future, it might be interesting to expand the present work to decimal numbers."

From previous work, "scientists already knew that fractions and whole numbers engage overlapping parts of the brain's number-processing network," Derbie said. "This study gives us a clearer picture of how their values are organized."

In the future, the team could examine how this neural representation varies across people of different ages, educational levels and mathematical backgrounds, Derbie suggested. The study also included only native French speakers, so may not reflect differences across language or culture.

"I would want to see how this organization develops as people learn," Derbie said. For instance, the team could follow children before and after they learn about fractions, to see whether changes in brain activity accompany better understanding.

In the current study, the participants found it more difficult to compare fractions than to compare whole numbers, the researchers noted; they were slower and made more errors while working with fractions, which was expected. But the study's results don't offer hints as to why.

"Our study was not designed to explain why fractions are harder to learn or understand," Valério and Dehaene told Live Science. "What is new in our study is that we show that these different types of numbers are organized within a shared map in the brain."

See how much you know about the most complex organ in the human body with our brain quiz!

'> Brain scans reveal 'number line' that's used to mentally compare values

Dwarf reed snakes are secretive, often nocturnal species that live across Southeast Asia, where they are prey for birds and other snakes. Although the snakes are extremely cryptic and usually hidden beneath leaves or other natural objects, people occasionally observe the serpents' dazzling rolls.

In a 2023 study, researchers formally documented the rolling behavior for the first time in one dwarf reed snake species (P. longiceps), which researchers observed in northwestern Malaysia in 2019. Kojima's team started independently observing the behavior around the same time in the Malaysian state of Sarawak, on Borneo.

"I had never seen anything like it before, and I immediately wanted to learn more about it," Kojima said.

For the new study, researchers filmed the rolling behavior multiple times in three dwarf reed species ‪—‬ P. saravacense, P. collaris and P. longiceps — between 2019 and 2022. The team also studied three reed snakes from another genus (Calamaria), which didn't perform the behavior.

Kojima said their study builds on the 2023 research by quantifying the rolling movements, confirming the overall movement pattern that research described, and revealing further details, including how the snakes stabilize their heads.

Snakes aren't the only tumblers in nature. For example, some desert spiders can roll away from danger, as can mother-of-pearl moth caterpillars (Pleurotya ruralis or Patania ruralis). However, the snakes are unique in that they seem to have good control over their rolls.

Rolling movements typically destabilize the head and disrupt sensory-guided adjustment, especially at high speeds, according to the authors of the new study. In other words, it's hard for animals to control their bodies when they're rolling around. The snakes' technique of keeping their heads relatively fixed and often elevated off the ground likely counters what would otherwise be destabilizing effects.

In addition to studying speed and form, the researchers found that the snakes could move in a relatively straight line, readily moving along a "racetrack" that was only 8 inches (20 centimeters) wide. The researchers noted that the snakes often ended their tumbling routine by diving underneath nearby objects, so they seem to remain aware of their surroundings.

The rolling is a behavioral innovation that allows the snakes to overcome their physical constraints and pull off an otherwise impossible movement, according to the researchers.

"Behavioral innovation can give rise to surprising new abilities that we would not expect from an animal's appearance alone," Kojima said.

Researchers have speculated that the bizarre behavior could surprise and confuse predators, in addition to being a rapid means of escape. The authors of the 2023 study also hypothesized that rolling would result in a patchy scent trail, and because king cobras (Ophiophagus) use scent cues to locate their prey, this could make it harder for the predators to hunt dwarf reed snakes.

How much do you know about snakes? Test your knowledge with our snake quiz!

'> Watch bizarre snake 'cartwheel' away from danger at more than 12 feet per second