Anthropic Threat Report Warns AI Models Are Nearing the Threshold for Bioweapons Assistance
Anthropic's September 2026 threat intelligence report warns that its latest AI models can no longer be assumed to fall below the threshold for providing meaningful assistance in bioweapons development. The report also documented AI-assisted drone weapons and Russian cyber espionage operations.
Anthropic released its September 2026 threat intelligence report this week, warning that its most advanced AI models can no longer be assumed to fall safely below the threshold for assisting in biological weapons development.
The report, covering activity from December 2025 through August 2026, marks the first time a major AI company has publicly stated that its frontier models pose a meaningful risk in the bioweapons domain. Earlier versions of Claude were considered limited in this regard. The newer models are more capable, and that capability creates new risks.
The report documented five cases involving biological research with dual-use potential. In one case, a scientist affiliated with a military research institute used Claude to draft a grant application for gain-of-function research on the chikungunya virus. The proposed methodology involved techniques designed to enhance transmissibility and immune evasion. In another case, a researcher used Claude over several weeks to study avian influenza, focusing on mammalian adaptation and airborne transmissibility.
Anthropic said it could not determine in all cases whether the researchers intended harm, as the requests were framed as legitimate scientific inquiry. The company has implemented stricter restrictions on dual-use biological queries for its newer models.
The report also documented a Russia-linked operation that used Claude Code to build an autonomous drone swarm system designed to select targets, including people, and issue detonation commands without human involvement. Anthropic described it as a lethal autonomous weapons system.
A separate Russian state-linked actor used AI-driven workflows to automate cyber espionage, including modifying malware in real time to evade security detections.
Anthropic said it disrupted all identified operations, banned the associated accounts, and shared intelligence with government and industry partners. The company said it will continue publishing threat reports as part of its commitment to transparency.
