Houthis Tried Using Claude AI to Build Hypersonic Missiles, Got Blocked

According to Israel Hayom, citing a report by the Associated Press, Houthi-linked individuals in northern Yemen used Anthropic's Claude AI to attempt development of several advanced weapons, including a multi-configuration missile capable of hypersonic speeds and a warhead that would use mobile phone hardware for mid-flight course correction. Anthropic says the users were blocked after detection and never achieved operational capability, though they did carry out at least one failed guided-rocket test - and then returned to Claude to ask why it had failed. The company also reported that Claude Code was used in place of human software engineers to develop guidance, navigation, and control software for the weapons.

The incident has sharpened a debate about whether AI companies should be legally obligated to notify governments when they detect attempts to misuse their platforms for weapons development. Supporters of a mandatory reporting requirement argue that state and intelligence agencies are better placed than private companies to assess the security implications of such activity, track patterns across platforms, and take action. Opponents caution that mandatory disclosure raises questions about jurisdiction, the definition of misuse, user privacy, and whether such rules could push bad actors toward less scrupulous AI providers. An unaffiliated weapons expert told the AP that the Houthis likely lack the manufacturing capacity and technical knowledge to actually build hypersonic missiles at this stage, but appear to be working toward reducing their dependence on Iranian weapons and components. Anthropic did not identify the individual users, only noting the activity originated in Houthi-controlled northern Yemen.