SAN FRANCISCO, United States — Anthropic said on Thursday, September 10, 2026, that it had stopped five cases in which researchers used its Claude artificial intelligence models for work that could support biological weapons development.
The California-based company disclosed the activity in a threat intelligence report examining prohibited uses of its technology between December 2025 and August 2026.
Anthropic said it disrupted every operation described in the report and used the findings to strengthen its safeguards.
The company also shared information with government authorities and other technology businesses when appropriate.
Biological Research Raises Safety Concerns
Anthropic described biological misuse as one of the most serious dangers associated with advanced artificial intelligence models.
Without adequate restrictions, the company said, such capabilities “could have catastrophic consequences.”
The report detailed five cases involving scientists who used Claude for research that could contribute to biological weapons development.
Anthropic said it blocked the researchers involved.
Assessing such activity can be difficult because scientific information capable of assisting weapons research may also be valuable in developing “a vaccine or a cure for a disease,” the company said.
Jacob Klein, Anthropic’s head of threat intelligence, described the problem as “an incredibly nuanced situation” in an interview with The New York Times.
Requests are not necessarily presented as explicit attempts to construct biological weapons, Klein said.
Determining whether highly technical scientific work has a legitimate medical purpose or a potentially harmful application can therefore require further investigation.
The cases involved Claude Haiku, Sonnet and Opus models.
Anthropic said none of the reported activities involved Claude Fable or its Mythos-class systems, apart from one case of distillation — a technique in which a larger model is used to train a smaller one.
Cyberattacks, Surveillance and Weapons Software
The biological cases formed part of a broader collection of activity attributed by Anthropic to suspected state-sponsored organisations, criminals, spyware businesses, propaganda institutions and politically motivated individuals.
The reported uses ranged from fraudulent dating applications and schemes involving hotel Wi-Fi networks to surveillance systems intended to identify dissidents.
Anthropic also documented six cases in which Claude was used to create software connected to conventional weapons, including firearms, missiles, armed drones and bombs.
Some of the work involved targeting and control systems, the company said.
A cyberespionage group whose activities were consistent with public reporting about the Russia-based organisation Midnight Blizzard allegedly used artificial intelligence to monitor whether security systems had detected its malware.
Anthropic said the system would then modify the malicious code repeatedly until it avoided detection.
The company also identified activity associated with the hacking group ShinyHunters, an Iranian propaganda institution and actors connected to a Russia-based espionage campaign.
Anthropic further accused several Chinese artificial intelligence laboratories of attempting to reproduce Claude’s capabilities through distillation. The company said it disrupted the relevant accounts.
Claude was increasingly used for more than answering questions or advising human operators, according to the report.
In some cyber operations, artificial intelligence systems performed or coordinated several stages of an attack while people selected targets and reviewed the results.
Anthropic said it had incorporated lessons from the cases into its systems “to better prevent, detect, and disrupt these activities in the future.”
Other AI Companies Report Misuse
Threat reports have become a regular practice among artificial intelligence companies seeking to document how their products are being misused and what measures they have taken in response.
Google said on Tuesday that one person had attempted to use its Gemini model to obtain a “complete, step-by-step technical guide” for producing weaponised biological agents.
Both reports indicated that cybercriminals and state-backed groups were adopting artificial intelligence for increasingly complex operations.
Calls for Restrictions on Advanced AI
The findings emerged amid renewed public debate about the speed at which powerful artificial intelligence systems are being developed.
Senator Bernie Sanders, a Democrat, has introduced legislation that would temporarily suspend advanced AI development and prohibit artificial superintelligence.
Speaking on the BBC’s Newsnight programme on Thursday, Sanders said warnings of a potentially catastrophic effect on humanity required governments to “slow it down.”
President Donald Trump took a different position, citing international competition over the technology.
“If we don’t win AI, we’re going to be put in a very bad position,” Trump said on Thursday.
The Anthropic report followed a warning from one of the company’s leading safety researchers, who said there was a greater than 10 per cent chance that artificial intelligence “could kill all humans” within the next decade.
Jakub Pachocki, OpenAI’s chief scientist, also called earlier in the week for “voluntary slowdowns” in advanced development until adequate protections are established.
“I am concerned no one is prepared for the consequences of a continued rapid rise in machine intelligence,” Pachocki wrote.
OpenAI, the developer of ChatGPT, would continue working on safeguards, he said, but the industry would require broader action.
The warning also prompted an open letter to Andy Burnham, the British prime minister, seeking a multinational treaty governing the development of artificial intelligence.
The letter called for governments to begin consultations on the rules that should apply to artificial superintelligence.
