Updated
Updated · Harici · Sep 15
Anthropic Exposes 2,000-km Missile and H5N1 Misuse of Claude by State-Linked Networks
Updated
Updated · Harici · Sep 15

Anthropic Exposes 2,000-km Missile and H5N1 Misuse of Claude by State-Linked Networks

3 articles · Updated · Harici · Sep 15

Summary

  • Anthropic said state-linked and criminal users misused Claude from December 2025 to August 2026 for weapons, bio research, surveillance and influence operations, then had their accounts terminated and indicators shared with partners.
  • A Houthi-linked cell in northern Yemen used Claude across three missile programs—including a ballistic missile with a range above 2,000 km—and returned after a failed live-fire test to upload telemetry and troubleshoot guidance software.
  • Iran-linked actors used the model to track US warships and personnel, draft targeting manuals and build surveillance tools, while a Russia-based team developed autonomous kamikaze drone-swarm software using battlefield footage from Ukraine.
  • Anthropic also logged five biological-risk cases, including work to boost Chikungunya transmission and adapt H5 avian flu—described as carrying a 50% human mortality rate—for airborne spread in mammals.
  • The report adds to Anthropic's recent warnings that frontier AI can accelerate bio-weapon and cyber capabilities, broadening the threat picture from pathogen misuse to military and political operations.

Insights

Tech giants are quietly intercepting AI bioweapon plots, but how many malicious prompts are already slipping through the cracks?
Could the same AI built to cure diseases secretly be helping rogue actors engineer the next global pandemic?