A recent 154-page report from the AI firm Anthropic alleges that a small group of freelance developers based in Russia used the company’s Claude AI chatbot to assist in the creation of software for autonomous 'kamikaze' drones. According to the report, the software was designed to facilitate terminal guidance, target selection, and coordination among multiple aircraft operating as a swarm. Anthropic stated that the technology allows these drones to identify and crash into targets autonomously, removing the need for a human operator to authorize the final strike.

The findings were part of a broader investigation by Anthropic into how malicious actors have attempted to exploit its AI systems for state-sponsored surveillance, propaganda, and cyber operations. While the report identifies the developers as likely freelancers, it highlights the growing concern among AI developers regarding the dual-use nature of their technology. The ability to automate target selection in weapon systems represents a significant shift in drone warfare, moving away from traditional remote-piloted operations toward fully autonomous engagement.

Anthropic has not provided specific details on the identities of the developers or the exact timeline of the software development, but the report serves as a warning about the potential for generative AI to lower the barrier to entry for creating advanced military technology. The company maintains that it has implemented safety guardrails to prevent such misuse, though this incident underscores the difficulty of policing the output of large language models in a globalized digital environment.