Anthropic says it blocked state-sponsored attempts to use Claude for bioweapons research

Anthropic, a San Francisco-based AI company founded by Dario Amodei, claims to have detected and blocked state-sponsored attempts to use its Claude AI model for bioweapons research. The company identified at least five instances where suspicious actors circumvented safety protocols designed to prevent access to Claude's knowledge base on infectious diseases and lethal toxins.

However, Anthropic estimates that there were about 35 distinct efforts by state-funded bioweapons developers between November 2025 and September 2026 to obscure their research objectives. These attempts included studies on chikungunya virus transmission, bird flu adaptation for severe illness in mammals, and orthopoxviruses, which encompass smallpox and mpox.

The company acknowledges the difficulty in determining whether such research is intended for bioweapon development or medical purposes. Anthropic's response to the bioweapons scare involved upgrading its safety protocols, introducing stronger safeguards in the latest models, particularly the Claude Fable 5 version, to restrict access to dual-use biological research queries.

The company did not disclose the regions from which these queries originated, but it is known to block users in China, Russia, Iran, North Korea, and several other countries. Anthropic emphasizes that the misuse of AI to create bioweapons poses one of the most serious risks to AI models, and the implemented guardrails aim to prevent such scenarios where rogue nation-states could exploit AI for biological warfare.