Anthropic blocks potential misuse of AI for biological weapons research

Anthropic released a report Thursday detailing how it had blocked bad actors from using its artificial intelligence technology. The company said it stopped individuals from using its AI chatbot Claude for cyberattacks, surveillance, and research that could have led to biological weapons. Anthropic said it added stronger safeguards to Claude’s latest models to restrict biological research that could be used for making weapons.

Anthropic said it has a responsibility to disclose malicious misuse of its services. It urged governments and AI competitors to identify and prevent similar threats. This marked Anthropic’s third report since March 2025 on malicious misuse. The report was published two days after one of Anthropic’s researchers resigned, citing ethical concerns that AI had no boundaries and was a threat to humanity.

What were highlights from the report? The company said it disrupted dozens of instances of potentially malicious use of its AI between December 2025 and August 2026. The bad actors ranged from spyware vendors, to financially motivated criminals, to state propaganda institutions. Several attempted threats included research that could have led to biological weapons, Anthropic said. For instance, an unnamed individual outside the United States used Anthropic’s AI to research the highly contagious avian influenza, or bird flu. The research centered on the virus' adaptation to mammals. While Anthropic said it did not assert that the researchers intended harm, the company insisted the information could be dual-use in nature and could be used for weaponization purposes.

What’s the latest from Capitol Hill on AI? On Thursday, Sen. Josh Hawley, R-Mo. launched an investigation into OpenAI after its intelligence system went rogue and hacked into another AI company. Hawley said that as the chair of the Homeland Security and Governmental Affairs Committee, as well as the Subcommittee on Disaster Management, he would look into the incident.

Also on Thursday, Sen. Chris Van Hollen, D-Md., asked OpenAI to allow federal cybersecurity agencies to assess the safety and risks of its AI models.

Dig deeper: Read Daisy Sachoff’s report for more details on why an Anthropic researcher quit his job.