Mutation hotspots help 'friendly' viruses outmaneuver the bacteria in your gut

Could we harness their chameleon-like nature to treat infections when antibiotics don’t work?

Michigan State University

Facebook

X
BlueskyMessageWhatsAppEmail

 

image: 

Certain bacteriophages found in the human gut have mutation hotspots scattered throughout their genomes that help them modify key defense genes, researchers report.

view more 

Credit: Sundharraman Subramanian and Alaina Pabbathi, Cryo-EM Core Facility, Michigan State University

Every 15 minutes, someone in the U.S. dies of a drug-resistant superbug. A few decades from now, antibiotic-resistant bacterial infections threaten to become the leading cause of death worldwide, outpacing cancer.

In the race for a solution to the antibiotic resistance crisis, a century-old practice is attracting renewed interest. The treatment, called phage therapy, involves co-opting friendly viruses that kill bacteria but ignore human cells.

Bacteria can — and do — develop resistance to phages, just as they do with antibiotics. But unlike antibiotics, phages can evolve counter defenses of their own.

Now, researchers at Michigan State University have identified a counter defense used by a group of phages common in the human gut, called Enterobacteria phage T2, that helps them stay one step ahead of their bacterial hosts.

These phages have mutation hotspots scattered throughout their genomes that help them modify key defense genes, the researchers report.

In a study published Aug. 13 in the journal Nature Microbiology, they show that these mutation hotspots help diversify their progeny to employ different survival strategies, ensuring that at least some continue to infect and kill no matter what countermeasures their bacterial hosts throw at them.

“They’re essentially hedging their bets,” said co-author Chris Waters, a core faculty member in MSU’s Ecology, Evolution, and Behavior program.

“If we can harness these kinds of evolutionary tricks, we might be able to make more effective phage therapies in response to the antibiotic resistance crisis,” Waters added.

The idea of using phages in medicine isn’t new. Cocktails of phages have been used since the 1920s to treat dysentery, sepsis, pneumonia and other ailments, particularly in France, Poland and parts of the former Soviet Union.

Interest in phage therapy waned in the West after the discovery of penicillin and other chemical antibiotics in the 1940s. But now, with deadly microbes from MRSA to tuberculosis becoming resistant to more and more of these drugs, researchers are revisiting phage therapy to combat antibiotic-resistant infections.

When phages invade, they latch onto a bacterium and inject their genes into the cell. Once inside, they hijack the bacterium’s internal machinery and turn it into a virus factory, forcing their host to churn out new phages until the cell bursts and releases them.

To fend off these attacks, bacteria have their own tactics. The researchers were studying one such strategy — a system in the bacterium that causes cholera — when they noticed something odd. In previous work, they identified a set of genes in cholera that spot the DNA of invading phages and chop it up before the phages can take over. But interestingly, this anti-virus protection didn’t last for long.

First author Jasper Gomez conducted the work while earning his Ph.D. in the Waters lab in MSU’s department of microbiology, genetics, & immunology.

In their experiments, the researchers transferred cholera DNA encoding the protective system to E. coli, a bacterium that is easier to work with in the lab, and exposed the bacteria to phages. Before long, the engineered E. coli were under attack. In other words, the phages quickly devised a workaround to bypass their hosts’ defenses, allowing them to sneak in and hijack their victims’ cells anyway.

“Within a few hours, the phages always started to win,” Waters said. “We couldn’t understand why,” he added.

The researchers sequenced the DNA of the resistant phages and found that many had “typos” in a gene called agt, particularly in a region of repetitive DNA where the same letter, or nucleotide base, appeared multiple times in the gene sequence.

“When I saw the data, I thought, oh my gosh,” Waters said. The region resembled a type of mutational hotspot called a contingency locus. Well studied in other organisms but never shown in phages before, such regions of the genome are known to be places where the cell’s DNA copying machinery sometimes “slips” and makes mistakes, Waters said.

The result is that, each time new phages are produced, they aren’t producing exact genetic copies of their ancestor. Some of the resistant mutants gain an extra repeat unit in the agt gene, while others lose one, throwing off how the gene’s instructions are read.

The researchers found that the repetitive region accumulates mutations thousands of times faster than the rest of the genome.

While mutations are often harmful, this changeability can give phages an evolutionary edge, Waters said. By continually churning out new mutants, they increase the odds that at least some will carry a mutation that lets them evade or disarm their host’s ever-changing arsenal.

“This changes our understanding of how phages evolve,” Waters said. “Instead of hijacking their hosts to mass produce exact copies of themselves, they are actually using these mutation hotspots to make a zoo.”

Phages outnumber bacteria by around ten to one, making them the most abundant organisms on the planet. The researchers focused on a type of phage that lurks in the gut, where it specializes on E. coli bacteria, but phages can be found just about anywhere, from the sands of the Sahara Desert to the ice of the Arctic Sea.

Working with MSU microbial evolution expert Jeffrey Barrick, the team found hundreds of similar mutation hotspots scattered across the genomes of other phage species as well.

Next, the researchers are looking into whether these mutation hotspots give phages an edge in other situations, such as adapting to survive and exploit their bacterial hosts after a shift in the environment, or evolving to infect new types of bacteria.

In much of the U.S., the U.K., and elsewhere, phage therapy is still far from mainstream; regulatory hurdles make it available only as a last resort. In the meantime, Waters and other researchers at MSU are exploring potential applications beyond the clinic, to treat bacterial infections in everything from honeybees and crops to pets and livestock.

“MSU could be a great phage therapy center for veterinary and agriculture applications,” Waters said.

“We’re never going to be able to completely get rid of resistance,” he added. “But if we can better understand how bacteria protect themselves from phage infection and how phages fight back, we might be able to minimize it.”

This research was supported by grants from the U.S. National Institutes of Health (GM139537, AI158433, GM088344 and F31AI186463) and the National Science Foundation (DEB-1813069 and DEB-1951307).

CITATION: "Phage-encoded contingency loci enable bet-hedging against host defence mechanisms," Jasper B. Gomez, Jeffrey E. Barrick, Christopher M. Waters. Nature Microbiology, Aug. 13, 2026. DOI: 10.1038/s41564-026-02445-w  

---

Journal

Nature Microbiology

DOI

10.1038/s41564-026-02445-w 

Method of Research

Experimental study

Subject of Research

Cells

Article Title

Phage-encoded contingency loci enable bet-hedging against host defence mechanisms

Article Publication Date

13-Aug-2026

 

Scientists create first AI-designed viruses to fight drug-resistant superbugs

Copyright AP Photo/Petr David Josek

By Marta Iraola Iribarren

Published on 07/08/2026 - EURONEWS

Researchers in the United States have used artificial intelligence to create viruses, opening pathways for new drug development and raising biosecurity fears.

Scientists are using a new form of generative artificial intelligence to design specialised viruses that can hunt and kill harmful bacteria. This breakthrough could lead to a new generation of antibiotics designed to defeat drug-resistant "superbugs".

Using the AI model Evo 2, which can create new DNA, scientists at Stanford University created a series of viruses known as bacteriophages – microorganisms able to kill bacteria.

In lab tests, a mixture of 16 “exceptionally good” viruses designed by Evo 2 was able to rapidly kill E. coli bacteria that were already immune to natural phages.

E.coli is a group of bacteria that can cause gut infections and is increasingly resistant to available antibiotics.

Antibiotic resistance is rising worldwide, driven mainly by the mis- and overuse of these kinds of antibiotics, which cause bacteria to develop ways to survive them, making current medicines useless.

“If the bacteria gain resistance to a single phage, it’s game over for the medication,” said Brian Hie, chemical engineer at Stanford and co-author of the study.

“But if you have multiple genetically distinct phages in a mixture, it would be harder for the bacteria to develop resistance to the entire cocktail.”

The researchers noted that this technology could be used in the future to target other harmful bacteria, such as those that cause tuberculosis, or a common hospital-acquired infection (MRSA).

Model available for everyone

The researchers have made the Evo 2 AI model openly and freely available for everyone to download and use, which has raised safety concerns.

The study authors acknowledged that making the tool open source has raised discussions about safety and that one primary concern is that "bad actors" could potentially use modified versions of the tool to design harmful biological agents.

“As the authors highlight, this raises some serious regulatory and safety concerns, to say the very least,” said Simon Clarke, associate professor in cellular microbiology at the University of Reading in the United Kingdom, who did not participate in the study.

“While work of this nature is normally tightly regulated, it is reassuring that these scientists have shown further restraint in providing important guardrails, but there is no guarantee that every other scientist attempting to do something similar will be so careful,” he added.

He argued that naturally occurring pathogens currently pose a greater risk than AI-designed ones, as they are already easier to access and produce than to create new ones from scratch.

According to the authors, an advantage of AI-designed biology over natural evolution is the ability to build safety checks directly into the process.

Related

• Experts warn open access bio-data could help AI design dangerous pathogens

The future of biology AI models

The model, Evo 2, was trained on millions of natural genomes from all across the world, allowing it to learn the complex "grammar" and rules that make a DNA sequence functional.

As with other biology AI models, this dataset includes biological data such as genetic sequences and pathogen characteristics

Currently, no universal framework regulates these datasets, and while some developers voluntarily exclude high-risk data, researchers argue that clear and consistent rules should apply to all.

In February 2025, Evo 2’s team announced that they had excluded pathogens infecting humans and other complex organisms from their datasets due to ethical and safety risks, and to “preempt the use of Evo for the development of bioweapons”.

Earlier this year, more than 100 researchers across the world wrote an open letter arguing that while open access to scientific data has accelerated discovery, a small subset of new biological data poses biosecurity risks if misused.

“The stakes of biological data governance are high, as AI models could help create severe biological threats,” the authors wrote.

The researchers said that striking the right balance between openness and necessary security restrictions on high-risk data will be essential as AI systems become more powerful and widely available.

AI designs a novel E. coli killer

Stanford UniversityLinkedInWeChat

BlueskyMessageWhatsAppEmail

 

image: 

Brian Hie and Aditi Merchant examine a protein structure generated by Evo 2, an AI tool that can suggest genome designs. Lab tests of Evo 2’s designs for an E. coli killer exceeded expectations. | Andrew Brodhead

view more 

Credit: Andrew Brodhead

In science, the suffix “phage” means to devour. Thus, the name bacteriophage might conjure images of tiny creatures that “eat” bacteria. While that conception isn’t 100 percent scientifically accurate, bacteriophages are lethal bacteria killers nonetheless, and scientists are excited about engineering phage DNA as a path to new antibiotics.

And so it is that chemical engineer Brian Hie and bioengineering graduate student Samuel King came to study bacteriophage ΦX174 (pronounced “FYE-ex-1-7-4”). Hie, assistant professor of chemical engineering and the Dieter Schwarz Foundation Stanford Data Science Faculty Fellow, created Evo 2, a generative AI model that solves complex biological challenges by suggesting new DNA sequences. Given just the barest snippet of ΦX174 DNA, Evo 2 will go to work writing novel genomes of new phages that kill bacteria – in this case, the microbe E. coli, which can be deadly when infections turn serious.

Until recently, Evo 2 was mostly confined to the computer, but in a new paper, Hie and King have transitioned it to a real-world lab. Based on genomes written by Evo 2, they have synthesized nearly 300 novel phages and tested them for effectiveness against E. coli. They later narrowed that list to a relative handful of 16 exceptionally good E. coli-killing phages.

“In this case, we wanted the model to generate the entire genome end-to-end in a single left-to-right pass. We didn’t add anything,” Hie explains of the process that produced thousands of options for his team. “In lab tests, a few of Evo’s suggestions had higher fitness than the native ΦX174.”

Resistance-resistant antibiotics

Hie and King chose ΦX174 because its entire genome is less than 6,000 base pairs long. Compared to the 3 billion base pairs of the human genome, the ΦX174 genome is relatively simple. But its ability to kill bacteria nonetheless makes it an attractive test case for Evo 2’s design powers.

Why Hie would want or need more than one E. coli-targeting phage comes down to bacterial resistance, a common weakness of modern antibiotics. Eventually, after years of heavy use, bacteria evolve immunity to the medications that kill them. The antibiotics lose effectiveness.

“If the bacteria gain resistance to a single phage, it’s game over for the medication,” Hie said. “But if you have multiple genetically distinct phages in a mixture, it would be harder for the bacteria to develop resistance to the entire cocktail.”

Thus, this work might lead to a resistance-resistant antibiotic. Scientists imagine cocktails of genetically diverse bacteriophages working in tandem to counter microbial immunity. Similar approaches might be used to develop phages that target other harmful bacteria, such as tuberculosis, methicillin-resistant Staphylococcus aureus (MRSA), and Pseudomonas aeruginosa, a bacterium that is a leading cause of medically resistant infections acquired in hospitals, among others.

“We have a proof of concept in the paper, where we show that this cocktail of 16 phages rapidly overcomes resistance in E. coli that is immune to native ΦX174,” Hie said.

New tools

The technical challenges of designing entire genomes – creating them nucleotide by nucleotide and then transferring them into bacteria – are neither easy nor inexpensive. King, first author on the paper, led the project.

As short as the ΦX174 genome is, Hie explains, it is extraordinarily difficult to stare at a 5,400-character string and make sense of it gene-by-gene. King developed a computational framework to help the researchers evaluate the genomes and to narrow down the list of candidates from thousands to only those the team deemed most interesting for further exploration.

“One of the main parts of the design framework was figuring out what traits the genomes should have based on ΦX174 and related phages,” King explains. “The framework involved several key steps: generating genomes using Evo 2, evaluating options based on the design criteria, selecting optimal candidates, synthesizing them chemically, and then testing them in the lab to see which genomes worked best.”

“With the cost of DNA synthesis still quite high,” Hie explains, “Samuel’s framework helped us focus only on the most viable alternatives.”

Open source, open doors

As a proof of concept, Evo 2 has exceeded expectations. Sensing its potential in medical and biological sciences, Hie offers Evo 2 open source and free of charge. Anyone can download Evo 2 and design new genomes themselves.

The free availability of the tool has naturally raised discussions about safety and security. Hie is motivated by the tool’s vast potential benefits to health and humanity, and said that open availability is essential to expediting research and real-world results. While he acknowledges that modified versions of the tool could be used by bad actors, he notes that existing pathogens present a greater risk than potential AI designs because they are easier to access and produce and, unlike with AI, safety checks can’t be built into the process of natural evolution. In fact, he adds, AI-enabled tools like Evo 2 provide humans a powerful advantage against naturally occurring pandemics and improved defense options against man-made biological threats.

Next up, Hie is working with other researchers at Stanford and beyond to extend Evo 2’s reach while pursuing his own research to create other new bacteriophages. He is also looking at longer and more complex DNA, possibly even of small bacterial genomes that might lead to beneficial engineered microbes that produce useful chemicals, medicines, fuels, and more.

“The biggest open questions for me,” Hie said, anticipating future research, “are how do we get greater genetic novelty and how do we get greater controllability of the outcomes?”

“One of the most rewarding parts of this project is the creativity Evo 2 allows,” King said. “New doors in science are now open because of what we can do with these models.”

 

---

For more information

Contributing authors include Stanford graduate students Claudia L. Driscoll, David B. Li, Daniel Guo, Aditi T. Merchant, and Garyk Brixi; and Max E. Wilkinson of the Broad Institute of MIT and Harvard and Memorial Sloan Kettering Cancer Center.

This research received funding from the Arc Institute, the National Science Foundation, the Knight-Hennessy Graduate Scholarship Fund, the Fannie and John Hertz Foundation, and the Stanford Institute for Human-Centered AI.

Media contact

Jill Wu, School of Engineering: jillwu@stanford.edu

---

Journal

Science

DOI

10.1126/science.aec2657 

Article Title

Generative design of bacteriophages with genome language models

Article Publication Date

6-Aug-2026

LA REVUE GAUCHE - Left Comment: Search results for PHAGE

UCF researcher to support DOE project using AI to accelerate scientific discovery

Assistant Professor Haonan Ling will explore virtual solutions to biomanufacturing for the Department of Energy’s Genesis Mission, which aims to strengthen America’s energy industry and national security

University of Central Florida

Facebook

XLinkedInWeChatBlueskyMessageWhatsAppEmail

 

image: 

Assistant Professor of Mechanical and Aerospace Engineering Haonan Ling's contributions to Department of Energy's Genesis Mission will address problems with biomanufacturing by developing an AI digital twin (virtual replication). (Photo by Antoine Hart)

view more 

Credit: Photo by Antoine Hart/UCF

Scientific breakthroughs often require years of experimentation, testing and refinement before researchers can answer some of society’s most complex questions. To explore how artificial intelligence can help accelerate that process, Assistant Professor of Mechanical and Aerospace Engineering Haonan Ling is joining the U.S. Department of Energy’s Genesis Mission.

The nationwide initiative encourages interdisciplinary teams to develop new AI models and research workflows capable of addressing national challenges across fields such as advanced manufacturing, biotechnology, critical materials, energy and quantum information. 

“The Genesis Mission harnesses the collective strengths of the nation’s leading institutions across academia, industry, and government to accelerate the pace of discovery,” says Winston Schoenfeld, UCF vice president for research and innovation. “UCF’s participation reflects the expertise of our researchers and talented students, whose contributions will help shape AI-enabled scientific workflows and transform technological advances into real-world solutions that fuel American competitiveness.”

A New Approach to Creating Fuels and Chemicals

As interdisciplinary scientific challenges become increasingly data-intensive, applying human ingenuity to advanced technologies creates possibilities to solve long-standing industry issues. 

Ling’s contribution to the Genesis Mission will aim to address problems with biomanufacturing by developing an AI digital twin (virtual replication).  This digital twin predicts and optimizes bioprocess performance, enabling improved process monitoring, decision-making, and scale-up.

The project will also provide opportunities for UCF researchers at different stages of their careers to contribute to the work. Pinzhen Lin, who will begin a doctoral degree at UCF’s College of Optics and Photonics in Fall 2026, will join Ling’s research group and lead development of the project’s real-time sensor, including its characterization and performance benchmarking. Jirui Fu ’24PhD, a UCF mechanical engineering doctoral graduate and postdoctoral scholar in the College of Engineering and Computer Science, will also assist with the project.

“The challenge is that conventionally scaling up biomanufacturing is slow and prone to failure, creating a need for smarter tools to accelerate development,” Ling says. “If successful, this project could significantly accelerate the development and deployment of sustainable biomanufacturing for fuels and chemicals, making the process faster, cheaper and less risky.”

With industry-academic collaboration at the core of the Genesis Mission, Ling is working on the project with Kansas State University Assistant Professor Yian Chen, as well as the National Laboratory of the Rockies researchers Ajinkya Pal, Jason DesVeaux and Evan Komp.

A Mission With Many Benefits

Although the Genesis Mission is focused on accelerating scientific discovery, Ling believes the work has the potential to create benefits that extend beyond the research community. 

“The AI digital twin framework developed here has strong potential as a commercial platform that can be adopted across a wide range of industries, from energy to materials,” Ling says. “More broadly, it could lower the barriers for industrial partners to adopt bio-based processes, helping drive the transition toward a more sustainable economy.” 

For Ling, the research also represents an opportunity to see emerging technologies applied to real-world challenges. 

“As an early-career researcher, I feel very fortunate to lead and participate in a mission of this scale,” Ling says. “What excites me most is the opportunity to apply this technology to solve real-world problems, and to see how it can be integrated with the rapidly advancing field of AI.”

-----------------------------------------------------------------------------------------------------

This project will be supported by the U.S. Department of Energy Office of Science through the Genesis Mission, a Transforming Science and Energy with AI initiative.

World Bank warns developing countries to embrace AI or be left behind

AFP
August 4, 2026

The World Bank has called for governments in developing countries to embrace the use of AI to achieve better governance outcomes – Copyright AFP Martin LELIEVRE

The World Bank on Tuesday called on developing countries to embrace artificial intelligence technology tools to deliver better governance outcomes, warning that they risked being left behind if they failed to do so.

“AI has thrown developing economies a lifeline, and they should seize it,” Indermit Gill, chief economist of the World Bank Group, said as the organization launched its annual World Development Report.

“They do not need large models or big data centers to reap its benefits,” he added, advocating for the adaptation of lower-cost AI tools to local conditions to deliver results in the health, education, justice and agricultural sectors.

Advanced AI models — largely developed in the United States and China — offer the ability to quickly analyze data and automate many tasks that otherwise take skilled humans longer to do.

These AI models, however, require huge data centers and large amounts of complex computing power, using massive amounts of electricity and water — with implications for climate change.

“Developing economies today are in the midst of their weakest average growth performance in three decades,” said a World Bank statement accompanying the report.

“AI could significantly boost that performance before the end of the 2020s while delivering tangible benefits to people.”

The report calls for countries to use AI to “help extend otherwise costly medical, legal, educational, and agricultural services to underserved billions — doing in a decade what might otherwise take a century.”



– Shock after shock –



Lower-income countries have struggled through the 2020s, hit by a series of successive shocks that saw the World Bank earlier this year dub it a “lost decade” for their economic growth.

The Bank has lowered its 2026 global growth forecast to its lowest level since the pandemic, with the economic fallout of the Iran war battering countries around the world.

The shock has hit low-income and developing countries hardest, with Asia the worst-affected region.

The Bank’s new report advocates for developing countries to start working with localized AI tools and solutions now, and to invest in electricity generation and distribution; expand access to computing power; and improve the availability of local data.

“The window to get this right is narrow,” said Gaurav Nayyar, director of the report.

“AI presents a once-in-a-lifetime opportunity to solve problems that have resisted solutions for generations,” he added.

For the 6.8 billion people — 83 percent of humanity — who live in low-income and developing countries, AI tools will need to be adapted to meet their needs.

The report shares examples of AI applications in governance, such as to increase diabetes screening volumes in Bangladesh, or in reducing costs for Indian farmers through advanced weather forecasts.

The solutions, the report stresses, will need to meet people where they are.

“For example, AI solutions will need to be delivered through voice calls on basic mobile phones for those who cannot read or afford smartphones,” it says.

“Simply importing an AI model does not mean it will work well locally.”



– Stark warning –



The report calls for policymakers to also build public trust as they expand AI use.

“Improved public services and better learning outcomes in schools will reinforce trust — but if AI embeds bias in government decisions or erodes data privacy, that trust will be difficult to recover,” said the statement.

The report delivers a stark warning, too: “AI could widen gaps between countries, increase inequality within them, concentrate market power, weaken trust in public institutions, and create new risks for safety, rights, and social cohesion.”

And while risks to employment in developing countries are low at the moment, it warns that in the long run AI tools could cut off economic mobility by eliminating many of the middle-class jobs that enable it.

The report was written with the aid of several of the world’s most advanced AI tools, including offerings from OpenAI, DeepSeek, Google and Anthropic, according to a disclosure.

Trump admin to review ‘closed’ AI models before release: reports

AFP
August 4, 2026 

Founded in San Francisco in 2015 as a nonprofit research lab by Altman, Elon Musk and others, OpenAI burst into the mainstream with the launch of ChatGPT in 2022 – Copyright AFP MARCO BERTORELLO

The Trump administration on Tuesday met with tech leaders to finalize a security review process for advanced AI models before their release, though it will apply only to “closed” models, US media reported.

However, it remains unclear when — or if — the White House will release the details of the process or how it will be implemented and enforced.

According to Axios, the review process will only apply to closed models that are tightly controlled by their developers. Major closed model developers include OpenAI, Anthropic and Google.

By contrast, Meta and Nvidia are developing “open” models that can be downloaded and updated directly by users. Open models are made by competitors in China, including DeepSeek and Moonshot, as well as Mistral in France.

Tuesday’s meeting at the White House reportedly included OpenAI, Anthropic, Google, Nvidia, Microsoft and Meta.

The move to exempt open models is likely aimed at helping the United States stay competitive with China.

However, there is a “deeper problem,” according to Martijn Rasser, vice president for technology at the Special Competitive Studies Project, a think tank founded and chaired by former Google chief executive Eric Schmidt.

“The United States still lacks a statutory, predictable process for evaluating the security of frontier AI models,” Rasser wrote in a LinkedIn post on Tuesday, adding that a voluntary framework applying only to closed models “concentrates the uncertainty” on a handful of companies.

– Rogue models –

The rapid advancement of artificial intelligence has created a sense of urgency in the United States to develop regulations that mitigate and prevent risks associated with the technology.

US President Donald Trump (C) and OpenAI CEO Sam Altman (L) attend an AI leaders’ meeting during the G7 summit in Evian, France, on June 17, 2026 – Copyright AFP Ludovic MARIN

However, the Trump administration has favored a light-touch, deregulatory approach to most industries, including tech.

In June, Trump signed an executive order that called for major AI developers to submit new models to the government for review 30 days before they are publicly released. It also gave the federal government a 60-day deadline to finalize a proposed framework.

That deadline passed on August 1 without any public announcement.

Recent cyberattacks carried out autonomously by software from OpenAI and Anthropic have raised concerns about the capabilities of advanced AI models.

On July 21, OpenAI confirmed that its software escaped a testing environment and attacked another company, Hugging Face. About a week later, it said the models had also targeted three additional companies.

Then on July 30, Anthropic revealed that it also found three incidents where AI models being tested “gained unauthorized access” to unnamed organizations.

Trump urged a balanced approach that addresses AI security risks while maintaining US competitiveness.

“We have to be careful in both ways. We don’t want to restrict them where all of a sudden we come in second to China,” he told reporters in the Oval Office last week.

In recent months, DeepSeek and Moonshot have developed their own powerful new models that have reignited fears about US competitiveness in AI.

Taiwan probes 17 China-funded firms over high-tech talent poaching

AFP
August 5, 2026 

Chips, also known as semiconductors, are vital for every electronic device, from smartphones to electric cars, and control of supply chains has become a major priority for the world’s biggest trading blocs – Copyright AFP/File JENS SCHLUETER

Taiwanese investigators said Wednesday they have raided dozens of locations as part of a probe into 17 Chinese-funded companies suspected of trying to “illegally” poach Taiwan’s high-tech talent to erode the island’s competitive edge.

Taiwan is a powerhouse in the global semiconductor industry, producing nearly all of the most advanced chips for companies including Nvidia and Apple.

China is racing to develop the advanced chips used to power artificial intelligence systems, as it faces export restrictions imposed by the United States.

More than 300 investigators searched 64 locations and interviewed 114 people in recent weeks, the Ministry of Justice Investigation Bureau said in a statement.

Investigators said the companies allegedly sought to conceal the identity of their Chinese financial backers while establishing “unauthorised” operations or “illegally” recruiting talent.

As AI competition intensifies, Taiwanese firms face the “challenge of having their core high-tech talent targeted and recruited by ‘Chinese’ companies, leading to the loss of their core competitive advantages,” the statement said.

Taiwan has long accused China of carrying out espionage activities on the island, which Beijing claims is part of its territory and has threatened to seize by force.

Last month, Taiwanese prosecutors charged a former TSMC employee with stealing the chipmaking giant’s “core” technology and attempting to leak it to China.

Prosecutors also have detained seven people, including an Nvidia worker, in a separate case involving the alleged smuggling of the US tech giant’s AI chips to China.

When rogue AI launches a cyberattack, who is legally responsible?

AFP
August 1, 2026

In July 2026, two OpenAI models undergoing testing left their confined environment — a scenario the developers had not anticipated – Copyright AFP Martin LELIEVRE

Recent cyberattacks carried out autonomously by two rogue OpenAI artificial intelligence models raises an untested legal question: who is responsible when AI acts on its own?

On Friday, Clement Delangue, head of the Hugging Face platform targeted by the intrusions, said there should be a way to “keep the companies that are doing some mistakes leading to (cyberattacks) accountable,” while saying his company would not be pursuing legal action at this time.

In mid-July, two OpenAI models undergoing testing left their confined environment — a scenario the developers had not anticipated — and ventured onto the internet, where they attacked Hugging Face, an AI model-hosting platform.

Delangue also mentioned Anthropic, which revealed Thursday that three of its models had broken into three different websites, also during testing.



– Negligence route –



Under US civil and criminal law, unauthorized access to a computer system is an offense.

“If a human OpenAI employee had broken into Hugging Face’s systems… OpenAI would be liable for the employee’s wrongful conduct,” University of Houston law professor Gabriel Weil wrote in an opinion piece for the Transformer newsletter.

“When an AI agent does it, the law treats it very differently, at least for now,” he added.

Matthew Tokson, a University of Utah law professor who focuses on new technologies, had a similar view, saying “we haven’t had to grapple with that being formed in anything that’s not human, and I don’t think courts are likely to be there yet.”

The question remains open, however, when it comes to the company that created the model.

“Does ‘we didn’t tell the AI to do that’ end the liability question?” asked Rob T. Lee, head of research at the SANS cybersecurity training institute, in a post on X.

University of Washington law professor Ryan Calo does not believe a criminal case would be likely to succeed.

“The company or individual would have to be at least reckless,” he said, explaining they would “be substantially certain the crime would occur and build or prompt the system anyway.”

Experts see greater potential for a civil — rather than criminal — case, where the burden of proof is lower.

“Some people think that AI companies should be strictly liable if an AI agent that they deploy totally breaks out, causes damages,” Tokson explained.

“Others would prefer to do like a negligence assessment and see if they were actually negligent or if this was just sort of an unavoidable accident or something that couldn’t possibly have been foreseen,” he added.

In such cases there is a standard of care in product design that judges or juries can use to make a ruling, Tokson continued.

“It’s all a bit unwritten because we’ve never had an AI agent break out of its sandbox and hack other people on the internet before,” he said.

OpenAI could rely on the lack of legal precedent if it faced a lawsuit, but those that follow will no longer be able to do so, Calo warned.

Proving that a similar incident could have been anticipated “shouldn’t be so hard now that it’s begun to happen.”

AI keeps consumer prices high in ‘RAMaggedon’ chip crunch

AFP
August 1, 2026

Copyright AFP Peter PARKS

Print-outs of articles about the global memory chip shortage are pinned beside a price list at a Hong Kong computer shop, offering an explanation to confused customers feeling the pinch.

Price rises for goods such as laptops and smartphones, with cars potentially next, have been an unwelcome side-effect of the artificial intelligence gold rush — and the squeeze is far from over.

Samsung Electronics’ chief financial officer said this week shortages of microchips that store digital data will likely deepen in 2027 and stay tight through 2028.

The crunch has been nicknamed “RAMaggedon” after the components called RAM, or “random-access memory”.

It is caused as profit-hungry chipmakers pivot to producing high-bandwidth memory (HBM) — a more advanced type of computer memory in huge demand to help train and run AI tools.

The articles on display at In-Technology Services — one of many compact vendors crammed into Hong Kong’s Wan Chai Computer Centre — are to inform customers who “don’t know what happened,” manager Wade Lam told AFP.

The centre’s shops sell tech equipment of all sorts, from computer parts to gadgets and games consoles.

Ken Tam, manager of Videocom Computer, which specialises in custom-built PCs, said business has halved since price rises began in September.

Sixteen gigabytes of RAM used to cost HK$300-400 ($40-50) but the price has now hit HK$1,500, he said.

“When it suddenly gets so expensive, customers have a psychological barrier,” Tam told AFP.

“If they need it, they will buy it,” but otherwise they will wait, or “lower their standards” and buy a less high-performing memory chip, he said.



– Chinese competition –



Analyst Ellie Wang at the Taiwan-based market research firm TrendForce said memory prices for PCs and smartphones were up around five to six times compared to a year ago.

The AI boom has brought humungous profits and share price jumps to the world’s top three memory chip makers: South Korea’s Samsung Electronics and SK hynix, along with US giant Micron.

In fourth place is ChangXin Memory Technologies (CXMT), which became mainland China’s most valuable company on Monday when it made its market debut in Shanghai — another sign of how red-hot the sector has become.

CXMT, as a relative newcomer, “remains in a follower position regarding leading-edge technologies”, James Zhao, senior principal analyst at Omdia, told AFP.

HBM is used in data centre servers to support other powerful chips — such as those made by US titan Nvidia — that execute the dizzyingly complex calculations of AI systems.

But when it comes to conventional RAM, and a type for computers called DRAM, “the current supply-constrained market environment” could bring CXMT “late-mover advantages”, he said.

At a shopping centre in a different part of Hong Kong, customer Henry Wong, an investment banker, said he had chosen to upgrade the RAM in an older laptop instead of buying a new one with even better specs.

“After upgrading the memory, I found it ran really smoothly, and I stopped wanting to buy a new computer,” he told AFP.



– ‘Bubble’ –



Automakers say they are facing rising costs for in-vehicle computer systems, which could also soon push up the price of new vehicles.

In Tokyo’s tech hub of Akihabara, Charles Brousse, a 30-year-old graphic designer and custom PC builder from Belgium, said prices for RAM, graphics cards and motherboards have hit “ridiculous levels”.

For his “PC & Chill” service, Brousse does not buy parts to pre-build machines — as it is too expensive — but he requires clients to purchase their own that he assembles.

The chip shortage is pushing people to buy cheaper laptops than desktops, which can last up to a decade, said Brousse, in Tokyo on his honeymoon.

“I’m not sure that’s a good thing; people end up buying products with shorter lifespans, which fuels a cycle of consumption.”

Brousse added that the fact it is driven by the “speculative bubble” of AI is also frustrating, “because I’m a graphic designer by trade, so AI has a real impact on my profession.”

Google rolls back new satellite image AI tool after backlash

AFP
July 31, 2026

This April 18, 2017 file photo shows people viewing a Google Earth map of Paris, France at an event at New York’s Whitney Museum of Art – Copyright GETTY IMAGES NORTH AMERICA/AFP SPENCER PLATT

Google rolled back Friday a new feature allowing Google Earth users to generate AI visualizations on top of the service’s satellite imagery, following a furious backlash from researchers and open-source intelligence experts about the potential for disinformation.

“We’ve seen geospatial professionals using this feature for a range of useful purposes, however we’ve also seen people sharing screenshots of generated imagery that appear to violate our policies,” a Google spokesperson said in a statement to AFP. “So we’re rolling back this feature in Google Earth while we work on implementing stronger guardrails.”

“We know that people uniquely trust Google Earth for a reliable view of the world,” the statement added.

The “create image” tool, unveiled Thursday, had integrated Google’s Nano Banana 2 image-generation technology into Google Earth, letting users zoom into locations and build pictures in seconds based on the program’s satellite, aerial and 3D-mapping data.

Google had described it as a way to “visualize history, create real estate plans and more.”

But disinformation researchers warned that it could enable bad actors to misuse Google Earth — a key resource for fact-checking and verification — to create realistic geospatial fakes.

“Google spent 20 years building the reference the world checks against,” digital investigations expert Henk Van Ess wrote in one widely shared post on Substack. “Today it added a button that makes things up.”

Brady Africk, a research analyst at the American Enterprise Institute specializing in satellite imagery analysis, told AFP the Google Earth update would have made it “easier to generate convincing fake satellite imagery that can spread quickly online and mislead the public.”

Such fakes “erode public trust in satellite imagery, and make the jobs of journalists and researchers more difficult,” he said.



– ‘Irrevocably damaged overnight’ –



As the critiques piled up, Google had said on X that the images created with the feature included SynthID digital watermarks, invisible markers embedded in media generated using the company’s artificial intelligence tools.

“We take misinformation seriously,” the company wrote, linking to an AI policy that instructs users against “misinformation, misrepresentation or misleading activities.”

AFP’s tests of the feature on Friday were able to fabricate satellite images that could have had serious global consequences: an explosion in Paris, a nuclear site in Iran, a bomb crater in Russia and an Islamic State group training ground in Syria.

The tool also fulfilled prompts to visualize other fake scenes, including floods across Bangladesh, a migrant caravan camped at the US-Mexico border and a suspicious warehouse holding vans in Georgia — one of several states US President Donald Trump has falsely accused of widespread election fraud.

The “create image” button no longer appeared within Google Earth after Google paused its use.

Similar fakes have previously had real-world impacts. In 2023, an AI image depicting an explosion at the Pentagon briefly rattled the markets.

Early in the US war with Iran, an AI-altered satellite image depicting a destroyed US military base garnered millions of views across platforms.

Among the groups that called for Google to reverse its decision was the London-based Centre for Information Resilience, whose executive director Ross Burley told AFP the feature was “irresponsible.”

“Trust in satellite imagery has taken decades to build and could be irrevocably damaged overnight,” Burley said.

GeoConfirmed, a volunteer-driven open-source intelligence project that verifies visuals from conflicts around the globe, had also urged Google to reconsider.

“We are already observing individuals exploiting the new Google Earth web tool to make military positions, relabel buildings as schools, and experiment with manipulating the information environment,” the group wrote on X, before Google announced it was rolling back the feature.

“The primary, and perhaps only, practical use of this tool appears to be the creation and dissemination of mis- and disinformation.”

NSF CAREER Award will help University of Illinois researcher improve the quality of AI training data

University of Illinois School of Information Sciences Assistant Professor Jiaqi Ma's project will develop new tools to understand how training data affects large AI systems

University of Illinois School of Information Sciences

LinkedInWeChatBlueskyMessageWhatsAppEmail

Assistant Professor Jiaqi Ma has received a National Science Foundation (NSF) CAREER award to develop new tools to understand how individual components of training data affect the behavior of large artificial intelligence systems. The highly selective CAREER award recognizes early-career faculty with the potential to serve as academic role models in research and education and to lead advances in the mission of their department or organization. Ma received a five-year, $660,307 grant for his project, "Data Attribution and Curation for Web-Scale AI Systems," which seeks to improve data attribution, a family of methods that estimate how training examples influence model behavior.

AI systems increasingly influence how people search for information, receive recommendations, learn, work, and create. The quality of these systems depends heavily on the data used to train them. However, modern training data is often enormous, noisy, and constantly changing, because it is drawn from many sources. Poorly understood data can reduce accuracy, amplify harmful information, weaken reasoning and/or make it difficult to recognize the value of content in AI systems. 

"Data is one of the most important ingredients in modern AI," Ma said, "but we still have a limited understanding of how individual training examples shape a model's behavior."

Ma hopes to develop tools to help researchers and practitioners decide what data to keep, remove, prioritize, or compensate for. By making data curation more principled and transparent, the project has the potential to improve the performance, reliability, and safety of widely used technologies such as language models and recommendation systems.

"This project aims to make those influences measurable, even in large and continuously evolving AI systems," he said. "Better data attribution can help us build models that are more accurate, reliable, and safe, while also creating a more transparent basis for recognizing and compensating the people whose content contributes to these systems."

The project will also create public educational resources, open-source software, conference tutorials, course modules, and research opportunities for graduate, undergraduate, and pre-college students, helping widen access to data-centered AI research and training.

Ma's research interests lie in the broad area of machine learning and AI, with recent focuses on the data foundations of AI, including three complementary aspects: 1) understanding how training data impact AI models (data attribution); 2) developing data-centric algorithms that improve the quality and safety of training data (data curation and synthetic data generation); and 3) studying how data mediate the societal impact of AI (data compensation and machine unlearning). His work has been recognized with a Best Paper Award from the Workshop on Navigating and Addressing Data Problems for Foundation Models at the 2024 International Conference on Learning Representations, and a New Faculty Highlight at the Association for the Advancement of Artificial Intelligence 2025.

He earned his PhD from the University of Michigan and worked as a postdoctoral researcher at Harvard University.

Finding new osteoarthritis medicines via AI and genetics

University of Utah Health

Facebook

XLinkedInWeChatBlueskyMessageWhatsAppEmail

 

image: 

Michael Jurynec, PhD

view more 

Credit: Kristan Jacobsen Photography / University of Utah Health

Key points:

• Researchers combined family genetic studies with AI molecular biology to find a new candidate drug to treat osteoarthritis.
• In cells in a dish, the new compound prevents OA-associated changes and promotes cell health.
• The drug has not yet been tested in people or animals.

IMPACT: The new drug is a starting point for therapies that treat OA at its source.

Osteoarthritis (OA) is a chronic, painful joint disease and a leading cause of disability. Despite its prevalence, therapies for osteoarthritis are limited and focus on symptom management. Now, researchers are combining genetic studies of Utah families with AI-based molecular biology tools to find new medications that may ultimately help treat OA at its source.

One such new drug appears to promote joint health and reduce inflammation-related genes in a model of osteoarthritis based on cells in a dish. While it has yet to be tested for safety and efficacy in a living organism, the new compound provides a starting point for innovative OA therapies.

“Our goal really comes down to treating patients,” says Michael Jurynec, PhD, associate professor of orthopedic surgery at University of Utah Health and the senior author on a paper describing the new results. “Right now, the only thing we can do for OA is joint replacement or pain medication. So, if we can find something that slows down the disease process, giving people an extra 10 or 20 years of pain-free living, that’s a huge advancement.”

The results are published in ACS Omega.

Finding new medicines

Using AI tools, the researchers were able to narrow a pool of half a million drug candidates down to six in a matter of weeks.

Previous human genetics research with Utah families had found that, for several forms of highly hereditary OA, changes in a gene called WNK2 underlie the disease’s progression. For these families, WNK2 overactivity in joint cells triggers processes associated with inflammation, which suggests that blocking WNK2 could effectively treat arthritis.

The scientists used an AI-based tool to predict the physical structure of the WNK2 protein, and then computationally simulated how hundreds of thousands of individual chemical compounds would interact with it. This gave them a “shortlist” of just over 50 compounds predicted to bind to WNK2 and reduce its activity. Visual inspection of the shortlist narrowed down the candidate pool to six compounds.

Promisingly, one of the candidate drugs, M04, appeared to prevent osteoarthritis-related changes and make cells healthier in an established cell-based model of osteoarthritis, in which human cartilage cells are exposed to conditions that trigger inflammation.

“We treated cells with this new compound we discovered, and it inhibited many, many genes that are associated with osteoarthritis,” Jurynec says. “Not only did it inhibit these inflammatory factors, but it actually increased expression of genes that promote the health of these cells.”

This suggests that the drug or a derivative of it could help treat OA.

Next steps

The new compound is a valuable starting point, but much more work lies ahead to develop it into a safe and effective drug, Jurynec emphasizes. While computational evaluation and in vitro studies suggest that M04 is a promising OA drug, its toxicity and side effects have not been fully tested, and researchers don’t know whether it will be safe to use in people.

The team is working to address these concerns through a collaboration with the University of Utah Therapeutics Accelerator Hub, working together to develop improved derivatives of the drug. M04 will also need to be comprehensively tested for safety and efficacy in animal models before clinical trials are possible. But as early as it is, the new compound provides a crucial starting point for development of better OA drugs.

“This is really the beginning of the study,” Jurynec says. “It’s not the end. We don’t have a drug that’s going to cure OA yet. But this is very promising.”

 

###

The results are published in ACS Omega as “Identification and Validation of a Novel WNK2 Inhibitor: A New Genetically Informed Target for Osteoarthritis Drug Development.”

This research was funded by the Skaggs Foundation for Research, the Utah Genome Project, and the Arthritis National Research Foundation. Content is solely the responsibility of the authors and does not necessarily represent the official views of the funding organizations.

Jurynec and first author Shivakumar Veerabhadraiah have filed a U.S. Patent Application (No. 19/672,326) titled ‘Compounds That Inhibit WNK2 Activity And Methods For Treating Osteoarthritis.’

---

Journal

ACS Omega

DOI

10.1021/acsomega.6c04569 

Method of Research

Experimental study

Subject of Research

Cells

Article Title

Identification and Validation of a Novel WNK2 Inhibitor: A New Genetically Informed Target for Osteoarthritis Drug Development

Article Publication Date

6-Aug-2026

COI Statement

Jurynec and first author Shivakumar Veerabhadraiah have filed a U.S. Patent Application (No. 19/672,326) titled ‘Compounds That Inhibit WNK2 Activity And Methods For Treating Osteoarthritis.’