Anthropic Reports Blocking AI Misuse for Biological Weapons and Cyberattacks
Anthropic released its third safety report since March 2025, detailing efforts to block malicious AI usage including biological weapons research and cyberattacks. The report highlights a blocked request for gain-of-function research on the chikungunya virus and notes that newer models like Claude Fable 5 have stronger safeguards against dual-use queries.
Anthropic has announced that it successfully blocked attempts by bad actors to misuse its artificial intelligence models for malicious activities, including cyberattacks, surveillance, and biological weapons research. This disclosure comes as part of the company’s third report since March 2025, which covers findings from December 2025 through August 2026.
Among the specific incidents detailed in the report, Anthropic systems prevented a request for assistance in authoring a grant application for gain-of-function research on the chikungunya virus. The proposed research aimed at enhancing the virus's transmissibility and immune evasion capabilities. The company noted that none of the reported cases involved the newer Claude Fable or Mythos-class models, with the exception of one illicit distillation case described as an industrial-scale covert campaign.
The report emphasizes that Anthropic has implemented stronger safeguards in recent models, such as Claude Fable 5, to restrict access to dual-use biological research queries. According to the company, older models were less capable of assisting sophisticated users in dangerous research compared to their newer counterparts.
In addition to biological threats, Anthropic identified nine cases of influence operations where groups created hundreds of social media accounts to amplify political views. These operations originated in various regions, including Russia, Iran, Turkey, the Persian Gulf, South Asia, Africa, and Europe.
The release of this report occurred two days after Anthropic researcher Jacob Coxon resigned. Coxon cited concerns that both Anthropic and OpenAI are not acting responsibly in AI development. The public disclosure of these novel threat activities is intended to urge government and industry action regarding AI safety.