Dustin Volz September 11, 2026 — 6:30 am Save You have reached the maximum number of items saved. Remove items from the saved list to add new ones. AAA Anthropic said it has foiled several potential plots this year by scientists who used its leading artificial intelligence models to conduct research that could help develop biological weapons. In a report describing abuses of artificial intelligence models, Anthropic said it could not determine whether the research served a legitimate or nefarious purpose, since valid biological research – the kind that can lead to breakthroughs such as vaccines – can also help create dangerous pathogens. In the face of this uncertainty, Anthropic said it was being cautious because the consequences of missing malicious activity could be severe. The potential for advanced artificial intelligence models to facilitate the development of known or entirely new biological pathogens is among experts’ biggest concerns about a technology that is advancing so quickly that even its leading architects doubt whether humans will be able to fully control it.AP “You don’t see in comics someone saying, ‘Hey, I want to create a bioweapon to kill everyone,’” Jacob Klein, head of threat intelligence at Anthropic, said in an interview. “It’s an incredibly delicate situation.” The potential for advanced artificial intelligence models to facilitate the development of known or entirely new biological pathogens is among experts’ biggest concerns about a technology that is advancing so quickly that even its leading architects doubt whether humans will be able to fully control it. Compared to the threat of catastrophic cyberattacks or AI agents that are inconsistent with human intentions, biological abuse has received less attention as an existential risk of AI, in part because past examples have typically only been demonstrated in research settings rather than in the real world. Andrew Weber, a senior fellow at the Strategic Risk Council who reviewed Anthropic’s report before its publication, said the findings are “chilling examples of how state-sponsored bioweapons developers are exploiting the rapidly evolving capabilities” of leading artificial intelligence models. A wide-ranging report from Anthropic published Thursday listed numerous abuses of artificial intelligence models, including chatbot Claude, over the past eight months. Some examples were similar to previous disclosures from Anthropic and other artificial intelligence labs, including suspected Chinese and Iranian government-linked individuals targeting dissidents and diaspora communities for surveillance. The report also highlights instances of Claude being used by Russian state media to create online propaganda masquerading as independent reporting, including fabricated claims about elections in Moldova. Anthropic documented another type of abuse that it said was new: attempts to use Claude to develop software for the design and development of conventional weapons, including firearms, missiles, armed drones and bombs. It details three cases in China, two in Russia and one in Yemen. The report does not identify by name which parties were involved in the Yemen case, but the context makes it clear that the Iranian-backed Houthi militias are involved. The use of AI by terrorist networks is a growing concern among US security officials. But among all the threat categories mentioned in the report, none is as worrisome as biological research. Anthropic has not disclosed the names of the researchers or institutions it blocked, their countries of affiliation or the specific biological agents in question, in part due to uncertainty about their targets. A wide-ranging report from Anthropic published Thursday listed numerous abuses of artificial intelligence models, including chatbot Claude, over the past eight months.AP Photo/Patrick Sison However, Anthropic said the scientists bypassed controls designed to prevent users in blocked regions from accessing its artificial intelligence models and worked to “obfuscate the purpose of their research in order to bypass our security measures.” The company has blocked accounts associated with the study. In one example from May, a scientist asked Claude for help writing a grant proposal to fund research to experiment with the chikungunya virus, a mosquito-borne disease that can lead to months of severe pain and other symptoms. Such research, known as gain of function, could be legitimate and lead to the development of vaccines, but it could also create superbug versions of viruses. In this case, the scientist wanted to create mutations of the virus that would make it more harmful as it repeatedly infected living animals. Anthropic said it believed the research was troubling in part because it could be understood that the work was intended to be performed at a military research institute. The findings are “a chilling example of how state-sponsored bioweapons developers are exploiting rapidly evolving capabilities.” Andrew Weber, Senior Fellow, Strategic Risk Council “What we don’t know is whether the research was intended to be used as a weapon,” Klein said. “But a military establishment conducting research to increase functionality is a concern.” Anthropic said evaluations last year of its older models showed they were not yet capable of significantly helping with dangerous biological research. The company says its current models are better equipped to handle complex scientific research, which has spurred tighter security measures aimed at limiting access to a wide range of dual-use biological research requests. Susan Monares, a microbiologist and public health expert who oversaw the Obama administration’s review of national biosecurity preparedness, also reviewed Anthropic’s report before its publication. She said the results provided real-world evidence to support fears that attackers are trying to secretly use advanced artificial intelligence “to improve their chances of creating biological pathogens that can cause significant harm.” Monares, who briefly served as director of the Centers for Disease Control and Prevention last year, said AI promises to usher in an era of medical breakthroughs that save and extend lives. But the same capabilities that make this possible, she added, could also “allow attackers to hide in plain sight, using seemingly legitimate research to create pathogens that we may not foresee and may not be able to stop once they are released.” Weber of the Strategic Risk Council, who also served as assistant secretary of defense for nuclear, chemical and biological defense programs during the Obama administration, called for restricting access to artificial intelligence tools capable of conducting this level of biological research to trusted researchers. “The fact that Russia, China and North Korea continue to develop banned biological weapons makes it imperative that we deny their researchers access to these extremely effective models,” he said. Save You have reached the maximum number of items saved. Remove items from the saved list to add new ones. From our partners Post navigation Will AI really kill us all? “Scary”: AI giant Anthropic says it has blocked possible attempts to create biological weapons