Anthropic report details Russian use of Claude for autonomous drone software
Anthropic's threat report says Russia-linked freelancers used its Claude AI to develop software for an autonomous combat drone swarm. The system could select targets and issue detonation commands without human intervention, and was trained on Ukrainian combat footage. Anthropic banned the accounts after identifying the activity as weapons development.
The DronDoc/Serafim project reportedly relied on Claude Code for swarm coordination, computer vision, and terminal guidance, with developers training the system on Ukrainian combat footage and simulating missions in Ukrainian locations. Hardware-in-the-loop testing was conducted on development boards, though field testing remains unconfirmed. The developers evaded Anthropic's geographic restrictions using commercial VPNs, and the company banned associated accounts after identifying suspected weapons development.
Beyond the drone program, Anthropic's report detailed a Russian state-linked espionage operation using Claude to automate phishing, malware development, and data exfiltration against over 20 organizations, including Ukrainian and European government and defense entities. Russia-linked actors also employed Claude for propaganda in the Central African Republic, producing pro-Russian and pro-Wagner content for radio, local media, and Telegram.
This report could signal a shift in how state-aligned actors develop advanced military technology, potentially lowering the barrier for autonomous weapons systems that operate without human oversight. If such tools become more accessible, the risk of accidental escalation or misuse may rise, particularly in conflict zones like Ukraine. The effectiveness of Anthropic's safeguards remains uncertain, and this case could prompt broader industry debate about AI providers' responsibility when their models are repurposed for harmful applications.