Anthropic blocks 'malicious use' of AI that could develop biological weapons

Anthropic announced that it has identified and disrupted attempts to misuse its AI systems for activities that could facilitate the development of biological and conventional weapons. According to the company’s latest threat‑intelligence report, the Claude model was employed in a Russia‑linked cyber‑espionage operation and by an Iranian propaganda outlet. The report details misuse detected over the preceding eight months, ranging from counterfeit dating apps and bogus Wi‑Fi hotspot schemes to surveillance tools aimed at locating dissidents. These findings follow earlier warnings from a senior information researcher at Anthropic concerning the broader risks AI poses to humanity. Anthropic also accused Chinese AI firms of attempting to replicate the capabilities of its Claude models. The threat‑intelligence report, released on Thursday, was published because the company believes it bears a responsibility to disclose any malicious misuse of its services. The document enumerates incidents involving suspected state‑sponsored actors, criminal groups, spyware vendors, government propaganda bodies and politically motivated individuals who exploited Anthropic’s technology. Specifically, “malicious use” of the Claude Haiku, Sonnet and Opus models was interrupted between December 2025 and August 2026. No misuse involved the Claude Fable or the Mythos‑class models, aside from a single case of distillation—the technique of training smaller models using larger, more expensive precursors. In addition to monitoring cyber and power‑operations, surveillance, scams, fraud and limb‑development applications, Anthropic said it blocked scientists who employed its AI in ways that could advance biological weapons programs. The report highlights five case studies in which users employed Claude models in manners that could support biological weapons development. Anthropic described biological misuse as “one of the most superior risks of frontier AI model,” warning that absent adequate safeguards such capabilities “could have catastrophic consequences.” The company added, “The same accusation that could be used to make a biological weapon could also be used to develop, for example, a vaccine or a cure for a disease.” Jacob Klein, head of threat intelligence at Anthropic, told the New York Times that the situation is “an incredibly nuanced situation.” He elaborated, “You are not seeing someone successful in a comic‑publication kind of way say, ‘Hey, I want to build a biological weapon to kill everybody.’” The report also cites six instances where Claude was utilized “to make bundle for conventional weapons, including firearms, missiles, equipped drones, bombs, and other munitions, as well as the targeting and power systems that run them.”