Anthropic has blocked attempts to use its Claude AI models for cyber attacks and biological weapons research, according to a report published Thursday. The company identified misuse between December 2025 and August 2026 by "politically motivated individuals, state-sponsored groups seeking to spread propaganda, and spyware companies."
One blocked request asked Claude to draft a grant application for research into enhancing the chikungunya virus. The mosquito-borne pathogen causes severe pain and fever, and the prompt sought funding to "create a new or enhanced biological property" through mutations that would make the virus more harmful. Anthropic said the request could have been "used to make the pathogen more dangerous."
Model capabilities drive safeguards
None of the documented misuse cases involved the company's newer Claude Fable or Mythos-class models, with one exception: an "industrial-scale, covert campaign to extract a model's capabilities and replicate them in another model without authorization."
The older models, including Claude Opus 4 and Claude Sonnet 4.5, "were well below the threshold" to assist with dangerous biological research. That meant safeguards focused mainly on preventing access to content that might help novices recreate known bioweapons. But Anthropic acknowledged the calculus has changed for current systems.
"For today's models - which are capable of assisting in a range of complex scientific research tasks - the evidence is no longer certain, and we cannot make that same assurance," the report said. The company has since implemented "stronger safeguards that restrict access to a wide range of dual-use biological research queries."
Researcher departure raises stakes
The report lands days after Jacob Coxon, an Anthropic researcher, announced his resignation. He said the company and competitor OpenAI "are racing straight to self-improving superintelligence and gambling with our lives." Coxon also warned that some colleagues believe Claude-class AI systems could threaten humanity by the end of the decade.
Experts have repeatedly urged governments to regulate AI directly as models grow more sophisticated, rather than relying on industry self-policing. The report's findings on biological research misuse add weight to that argument.
Why this matters for science and research professionals
Researchers working with dual-use technologies should expect tighter access controls on commercial AI tools. Anthropic's shift toward restricting a wider range of biological research queries signals that model providers are treating legitimate scientific prompts with greater scrutiny. If your work involves virology, synthetic biology, or pathogen studies, review your institution's policies on AI-assisted research before relying on external models for grant writing or experimental design. For broader context on how AI intersects with scientific work, see AI for Science & Research.
Your membership also unlocks:
