Anthropic released a threat report in September 2026 detailing how Iran-linked entities and Houthi groups utilized the Claude AI model for military intelligence and weapon development. The company identified and blocked 16 accounts associated with Iranian paramilitary and security organizations following internal investigations.
One incident involved using Claude to analyze open-source data on US naval forces in the Middle East, including aircraft and ship identifiers, satellite imagery, and movement patterns. The model also helped extract names of US military personnel from photo captions and analyze vulnerabilities in communication hardware to assist in target selection.
In Yemen, Houthi-linked groups employed Claude Code for three missile projects, including a multi-stage ballistic system with a range exceeding 2000 km and an R2000 variant featuring a hypersonic glide vehicle. The AI served as an engineering tool for software development, simulation, and control system tasks.
Iranian security units also used the platform for domestic surveillance, processing 155 216 posts on X to build profiles for 6388 opposition members over a year. Additionally, the AI assisted in creating digital tracking tools, such as a Firefox extension designed for mass data collection from social media platforms.
Anthropic has implemented enhanced detection systems to identify suspicious activity and shared its findings with government authorities. The company stated that the AI effectively performed tasks that would otherwise require a full team of engineers and programmers.
