Anthropic says it blocked misuse of its AI that could have supported biological weapons

Anthropic, an artificial intelligence company, announced on Thursday that it had thwarted attempts by malicious actors to exploit its AI models for illicit purposes such as cyberattacks, surveillance, and research potentially leading to biological weapons. As AI models continue to advance, they become increasingly capable of conducting complex cyberattacks and spreading misinformation.

In response, Anthropic has implemented stronger safeguards in its latest models to prevent misuse related to biological research that could be employed in creating dangerous pathogens.

In a comprehensive report released on August 9, 2026, Anthropic detailed nine instances of misuse identified since December 2025 and August 2026, involving actors ranging from state-sponsored groups to individuals involved in political propaganda. One particularly concerning case involved researchers attempting to use Anthropic's AI system, Claude, to support the creation of a more harmful version of the chikungunya virus, which causes severe pain and fever.

While such research could potentially aid in developing better vaccines, Anthropic emphasized that it could also be used to enhance the virus's harmful properties.

Anthropic clarified that none of the cases in its report were linked to its more powerful Claude Fable or Mythos-class models, released in 2026. The older Claude Opus 4 and Claude Sonnet 4.5 models, released in 2025, were deemed insufficient for assisting users in conducting dangerous biological research. Consequently, safeguards on these models were more lenient, primarily focused on blocking access to content that might help novices recreate known bioweapons.

However, Anthropic acknowledged that with today's advanced models, such as Claude Fable 5, it could no longer guarantee the safety of allowing unrestricted access to dual-use biological research queries.

Anthropic has taken action to strengthen its safeguards in response to these findings, sharing the information with government authorities and industry partners. The company expressed hope that its report would help other developers recognize similar patterns, enable governments and civil society to better understand emerging threats, and collectively enhance defenses against potential misuse of AI technology.