Yemeni militants used Anthropic AI to try to build ballistic missiles (www.ft.com)

Yemeni militants attempted to leverage Anthropic's Claude AI model to create guidance and control software for ballistic missiles, as revealed in a recent intelligence threat report from Anthropic. Although the report did not explicitly name the Iranian-backed Houthi rebels, it indicated they were part of a group in northern Yemen engaged in multiple weapons development programs. The use of AI in this context underscores the escalating risks associated with advanced technologies, as unauthorized actors may harness these tools for potentially harmful purposes. Anthropic reported that the militants replaced traditional software engineers with Claude, employing it to develop complex software for launching missiles and run simulations. This incident is significant for the AI/ML community as it raises concerns over the security and ethical implications of AI technologies falling into the hands of those with malicious intent. Anthropic's safeguards managed to block some requests, but the actors displayed resourcefulness by circumventing these measures, highlighting vulnerabilities in regulating AI applications. While no operational device has been confirmed to have been successfully fielded, the militants conducted test fires of a guided rocket and returned to Claude to analyze their failures. Anthropic responded by banning associated accounts and sharing threat intelligence, but the ongoing threat remains evident as the militants built offline simulation tools independent of the AI system.