Anthropic AI models misused for weapon development and cyber operations
Anthropic reported discovering that actors in China, Russia and Yemen misused its Claude AI models to develop software for conventional weapons including missiles and drones, as well as for cyber operations and surveillance. The company stated it found no evidence of successful operational weapon creation and blocked the misusing accounts.
What happened
Anthropic published a report identifying that its AI models were misused for conventional weapon development and cyber operations between December 2025 and August 2026.
The report documented use of Claude Haiku, Sonnet, Opus and Claude Code models for developing software for firearms, missiles, armed drones, bombs and targeting or control systems. Anthropic's threat intelligence team identified actors in China, Russia and Yemen using Claude to develop weapon design software or support intelligence gathering and procurement for weapons programs.
A group in northern Yemen used Claude Code to support three different weapon development programs involving guided rockets and various missiles, employing the system for directing weapons, developing navigation and balancing software, simulation and testing. Anthropic stated it found no evidence the group successfully developed an operational weapon.
Reported impact
The misuse involved multiple categories: cyber operations, influence operations, surveillance, fraud, and conventional weapon development. Perpetrators included state-backed groups, financially motivated criminals and state propaganda bodies.
Governance controls
Anthropic's threat intelligence team investigated and blocked activities of actors misusing the models. The company shared information obtained with public and private sector organizations.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.