Anthropic intervened this year to halt five potential operations in which scientists used the American company's leading artificial intelligence models for research that could contribute to the development of biological weapons. The cases are described in the report "Detecting and countering misuse of AI," released by the company, which warns about the increased capabilities of AI models in scientific research and the risk that these could be used for malicious purposes.
According to Anthropic, older models, such as Claude Opus 4 and Claude Sonnet 4.5, released in 2025, were "well below" the capability level from which they could provide significant assistance to a sophisticated user in conducting dangerous biological research. The situation changed with the more recent models, which are capable of supporting complex scientific research tasks. As a precaution, the company released the latest models, including Claude Fable 5, with stronger safeguards, restricting access to a wider range of dual-use biological research requests.
The company identified five specific cases. In one, a resale platform circumvented regional blocks to make the models available to virologists involved in a state-funded project researching the chikungunya virus. In another case, a researcher used Claude for several weeks to plan experiments related to the adaptation of avian flu to mammals. A third case involved a brokerage service through which Opus 5 produced a complete application for a research grant on orthopoxvirus immune evasion. In the remaining two cases, a state-backed researcher used the models to create a venom peptide atlas and a generative molecule optimization system, and another researcher used Claude to computationally redesign toxins as part of a national program.
Anthropic emphasizes that the cases do not allow the conclusion that the researchers necessarily intended to cause harm, since the same research may have legitimate applications. After detecting the cases, the company banned the associated accounts and incorporated the findings into its safety tools. Anthropic believes that disclosure of these cases should contribute to broader debate among AI companies and governments about the risks associated with using the technology in biological research.
