Iran and Houthis Exploit AI for Espionage, Military Aims, and Repression

AI developer Anthropic has revealed findings detailing Iran's use of its Claude AI model for intelligence gathering, military development, and internal repression between late 2025 and August 2026. One user, linked to the Islamic Revolutionary Guard Corps (IRGC), created an automated intelligence tool using Claude to collect and analyze open-source data on hundreds of Israelis and members of the Jewish diaspora in Western countries. This system helped identify potential targets and build detailed profiles of Israeli and American citizens, including government officials and NGO employees.

Another Iranian user employed Claude to devise strategies for attacking U.S. Navy vessels in the region, with the AI assisting in tracking ship locations, identifying personnel, and pinpointing vulnerabilities in maritime infrastructure. Anthropic views these instances as a significant escalation in AI capabilities, enabling complex operations from information gathering to hostile actions, though human oversight remains crucial for final target selection.

Separately, in northern Yemen, a group of engineers utilized Claude to aid in developing a ballistic missile with a range exceeding 2,000 kilometers and a hypersonic glide vehicle. After a failed missile test, the engineers used the AI to analyze the cause of the malfunction. Anthropic found no evidence that these missile developments reached operational status.

Within Iran, Claude was used for domestic surveillance. Two security units created 16 accounts to monitor social media posts and profile thousands of citizens over a year. One unit also developed a malicious browser extension disguised as a prayer schedule app to covertly collect user data from Iranian social networks.

Iranian propagandists leveraged Claude for "cognitive warfare," creating fake personas and disinformation networks, sometimes attributing false claims to Western research centers for credibility. A similar tactic was observed with the Iranian opposition group "Mojahedin-e Khalq," which used Claude to analyze thousands of posts to simulate a real activist. Additionally, entities linked to the UAE government used the model to prepare false testimonies for UN institutions.

An Israeli-Singaporean company attempted to build a platform for analyzing users in Iran and the Persian Gulf to identify political leanings and generate fake personas. However, this project was experimental, and Anthropic blocked access before it could be deployed against real targets.