Definition
The use of frontier AI systems — particularly AI coding agents — to develop, simulate, or optimize conventional weapons systems including guided rockets, ballistic missiles, and related guidance software.
Key Points
- 2026-09: gtg-87001 case is the first documented instance of claude-code substituting for human software engineers in guidance-navigation-control development (2026-09-13-anthropic-threat-report-september-2026-primary)
- Six conventional-weapons investigations in Anthropic’s September 2026 report span Yemen, China, and Russia
- Actors used parallel-agent-workflows, safeguard evasion, and compiled offline toolkits before bans
- Policy implications: export controls, AI lab safeguard gaps, embedded monitoring for coding agents
No operational deployed weapon confirmed in GTG-87001 case. Failed guided-rocket field test documented.
Related
- conventional-weapons-ai-misuse
- gtg-87001
- guidance-navigation-control
- claude-code
- ai-safety
- ai-governance
- generative-threat-groups
- parallel-agent-workflows