Anthropic announced that it has identified and stopped attempts to use its Claude AI model for activities that could aid biological and conventional weapons development, cyber‑espionage, and influence operations. The company’s threat‑intelligence report, released Thursday, details how its models were employed in fake dating apps, hotel Wi‑Fi scams, and surveillance aimed at dissidents.
According to the report, the misuse of Claude’s Haiku, Sonnet and Opus variants was intercepted between December 2025 and August 2026. No incidents involved the newer Claude Fable or Mythos‑class models, except for one case of model distillation. The firm also flagged five instances where its technology was used to facilitate biological‑weapon development, a risk it described as “one of the most serious” for frontier AI.
Anthropic highlighted six cases of the model being used to design conventional weapons such as firearms, missiles, armed drones, bombs and their targeting systems. Cybercriminals and state‑backed hackers, including the ShinyHunters group and China‑based labs, were named for leveraging Claude to build malware that could automatically evade detection.
The company said it had shared its findings with authorities and industry partners, and it has incorporated the intelligence into its processes to better prevent, detect and disrupt future misuse.
Anthropic’s chief of threat intelligence, Jacob Klein, told the New York Times the situation is “incredibly nuanced,” noting that the same information that could create a biological weapon could also be used to develop vaccines or cures.
Industry experts have warned of the rapid rise of AI, with OpenAI scientist Jakub Pachocki calling for voluntary slow‑downs and lawmakers like Bernie Sanders proposing legislation to pause advanced AI development. Anthropic’s report follows a top safety researcher’s warning that AI could pose a catastrophic risk to humanity within the next decade.
