Anthropic Disrupts Attempts to Misuse AI for Biological Weapons
Anthropic has thwarted several attempts to misuse its AI model, Claude, for potential biological weapons development. The company reported these findings, revealing efforts made by individuals over the past eight months to leverage AI for harmful research.
Sources
Anthropic has identified and disrupted attempts to misuse its AI model, Claude, for malicious activities that could support the development of biological weapons, with findings highlighted in its latest threat intelligence report published on September 11, 2026. Over the past eight months, instances of misuse included fake dating apps, scams, surveillance for identifying dissidents, and attempts to develop biological weapons, with five specific case studies presented in the report. The report also indicated that AI technology has been utilized by cybercriminals and state-sponsored groups, raising substantial concerns about the potential catastrophic consequences of AI misuse without proper safeguards.
Anthropic reported several attempts to misuse its AI models, including Claude, for research potentially linked to biological weapons over the past eight months. In one noted case, a scientist sought assistance to engineer mutations to the chikungunya virus for harmful applications, leading the company to ban involved accounts due to military research associations. As concerns about AI's potential dangers grow, Anthropic emphasized that newer models are capable of complex scientific work, prompting calls for restricting access to trusted researchers.
Anthropic AI reported thwarting several malicious operations using its Claude models, including an attempted missile guidance software development for a guided rocket and a long-range ballistic missile in northern Yemen, where it blocked various requests despite some slipping through. The threat report also highlighted state-linked cyber-espionage activities from Russian and Chinese groups, with the latter using Claude for offensive operations targeting government and corporate networks across multiple regions, while three Iranian accounts were banned for conducting covert influence operations. Anthropic's actions included banning involved accounts and enhancing monitoring to detect similar threats, along with addressing unauthorized access incidents following internal safety concerns.