ANTHROPIC said in a Thursday AI misuse-detection report that an unidentified
northern Yemeni group used its Claude model to support development of three
weapons projects: a guided rocket, a ballistic missile the group claims can
exceed 2,000 km range, and an R2000 missile series. The group used Claude to
develop guidance, navigation and control software—partially replacing software
engineers—and compartmentalized work across multiple AI sessions while
concealing the projects’ military nature to evade detection. The disclosure
arrives as Iran-backed Houthi forces continue missile and drone attacks on Red
Sea shipping and push into coastal areas near the Bab al-Mandeb, intensifying
regional fighting and adding upside pressure to energy markets.