Anthropic reported in September 2026 that a group based in northern Yemen utilized the Claude AI model to assist in the development of missile technology and associated software. The users employed the tool for coding, system configuration, and flight simulation related to guided and ballistic weapons.
The group worked on projects including a multi-stage ballistic missile with a range exceeding 2,000 km and R2000 rockets, which reportedly included a hypersonic glide vehicle variant. Anthropic noted that while the users conducted a test launch, it appeared to be unsuccessful, prompting them to consult the AI to diagnose the failure.
To bypass safety protocols, the users distributed tasks across multiple sessions and obscured their objectives. Anthropic blocked the associated accounts and shared threat intelligence with government and private sector partners. The company stated that the group managed to create an independent set of simulation tools.
Anthropic's September 2026 cybersecurity report also highlighted broader misuse of its models, including instances involving biological research. The company identified cases where users attempted to leverage AI for developing biological weapons or conducting research on virus function enhancement linked to military institutions.
