On Thursday, September 10, American artificial intelligence company Anthropic released its latest threat report confirming the banning of multiple user accounts located in northern Yemen. The report accuses these users in the Iran-backed Houthi armed control zone of utilizing their AI model Claude to replace the role of software engineers in developing precision missile and rocket systems.
According to Anthropic’s investigation, these users deployed Claude in three weapon projects, including a guided rocket, a multi-stage ballistic missile purported to have a range exceeding 2,000 kilometers (1,243 miles), and a hypersonic glide missile.
Operators used Claude Code to develop guidance, navigation, and control software to achieve automatic tracking and stabilization of aircraft. To evade security scrutiny, users segmented tasks into different conversations and concealed their military intentions in order to prevent a single prompt from exposing the entire operation.
The report notes that the group conducted a live test of a guided rocket and, following a launch failure, consulted Claude to diagnose the cause of the malfunction. Additionally, users developed an offline simulation toolkit to continue research and development in cases where AI connectivity was unavailable.
Anthropic stated that there is currently no evidence to suggest that the aforementioned research outcomes have been turned into operational weapons and deployed. Following the investigation, the company immediately banned the involved accounts, reinforced security measures, and shared threat intelligence with public and private partners.
In addition to the missile projects in Yemen, Anthropic’s report also uncovered hacker and intelligence operations linked to the Chinese, Russian, and Iranian governments.
A university student group in Hunan Province utilized Claude as a coordination layer to launch attacks on government networks in the Middle East, Europe, and Southeast Asia. Other accounts used AI to recruit and infiltrate Syrian Uyghurs using local dialects.
Groups associated with the Russian intelligence agency APT29 (Midnight Blizzard) automated processes using AI to conduct phishing and data theft against European, Ukrainian diplomatic targets, and drone manufacturers.
Official Iranian accounts used Claude to generate propaganda content and carry out covert psychological warfare operations.