Anthropic has reported that a Yemen-based weapons engineering cell utilized its Claude artificial intelligence model to develop guidance software for missiles and guided rockets. The group reportedly employed the AI to write control and position-estimation software, conduct flight simulations, and troubleshoot a failed rocket test. While the company found no evidence of successful weapon deployment, it noted that the actors attempted to circumvent safety protocols by concealing their objectives.
The Yemen operation was among six global activities disrupted by Anthropic, which also identified misuse involving drone swarms, torpedo interception systems, and electronic warfare software. The company stated that AI tools can effectively reduce the human engineering labor required for weapons development, potentially allowing less sophisticated actors to perform complex technical tasks. In response, Anthropic has banned associated accounts and shared intelligence with relevant partners.
Beyond conventional weapons, the report highlighted concerns regarding the potential exploitation of AI for dangerous biological research, including studies on bird flu and chikungunya. Anthropic emphasized that while its models can accelerate legitimate scientific progress, they pose significant risks if misused. The company is now implementing enhanced classifiers to detect and block activities related to weapons development and high-yield explosives.
