AI As A Weapon: Iran and Houthis Used AI to Track US Warships and Build Guided Missiles, Discloses Anthropic
Summarized by AI; it may make mistakes. Check important info
Summarized by AI; it may make mistakes. Check important info
Artificial intelligence developer Anthropic disclosed on Thursday that Iranian state-linked groups and Yemen-based Houthi militants used its Claude AI model to track American warships in the Red Sea and write guidance software for long-range ballistic missiles.
The San Francisco-headquartered AI firm published a 154-page threat intelligence report detailing covert operations disrupted between December 2025 and August 2026 across Iran, China, Russia, and Yemen. According to the disclosures, hostile actors deployed specialized tools, including Claude Code, in place of human software engineering teams to bypass technical bottlenecks and accelerate conventional weapons development.
Houthi Cell Used AI to Develop Missile Guidance Systems
In one of the primary cases, Anthropic identified a Yemen-based weapons engineering unit linked to Houthi operations that used Claude Code to write, test, and refine guidance, navigation, and control software for flying weapons.
The report revealed that the group deployed Claude to integrate an open-source autopilot system with a commercial, phone-class flight computer. The AI model was instructed to write control and position-estimation software, run firmware builds, and execute flight simulations for multiple missile projects. These included a guided rocket, a multi-stage ballistic missile with a projected range exceeding 2,000 kilometres, and a hypersonic glide vehicle variant known as the "R2000" set.
To evade automated safety protocols, the operators ran simultaneous Claude instances, assigning coding, research, and code review to separate sessions while masking the ultimate military applications. Anthropic confirmed that while the actors conducted a physical test launch of a guided rocket which failed there is no evidence that the group successfully fielded an operational AI-guided weapon.
Iranian State Agents Tracked US Naval Positions via Open Data
In a separate operation, Iranian state-aligned actors utilized Claude to gather open-source intelligence and construct precise targeting recommendations against US naval forces operating in regional waters.
The report stated that Iranian operatives fed Claude raw public data including satellite imagery query scripts, commercial vessel transponder records, military photo captions containing names of US personnel, and ship tracking databases to build structured targeting handbooks. Operators also instructed the AI to identify specific technical vulnerabilities in shipboard hardware, such as Cisco communications platforms, maritime satellite terminals, and industrial control systems.
Other Iranian operations used free Claude accounts to write custom domestic surveillance tools combining automated vehicle number-plate recognition with mobile device interception, alongside malware pipelines targeting dissidents and Israeli individuals
AI Collapses Engineering and Labour Gap in Warfare
The threat report highlights a structural transition in modern warfare, where generative models allow military groups to scale complex technical operations without expanding physical infrastructure.
"The cybersecurity skills of AI models means that AI has collapsed the labor and tooling gap that used to separate well-resourced, state-sponsored operations from individual operators," Anthropic noted in its assessment, emphasizing that "sophisticated attacks no longer require sophisticated attackers."
Anthropic confirmed it banned all user accounts associated with the rogue operations, updated its automated safeguard filters, and shared threat indicators with international security agencies and private sector partners.
