Anthropic flags AI-powered missile design, mass surveillance and recruitment schemes
Anthropic's latest threat report details how its Claude models were employed in missile development, large-scale phone surveillance and recruitment of Uyghurs, among other illicit uses.
Anthropic released its most comprehensive threat intelligence dossier, revealing that its Claude Haiku, Sonnet and Opus models were misused across seven harm categories between December 2025 and August 2026. The company identified operations by state-linked spies, weapons engineers, commercial spyware firms, propaganda outfits and ordinary criminals, then shut down the accounts and fed the findings back into its safety systems.
Highlights include an Iran-connected actor generating targeting data on US naval forces, a Yemen cell employing Claude Code to program guided rockets and a long-range missile, a PRC-aligned campaign profiling and attempting to recruit Uyghurs in Syria, and a Malian consultant using Claude to build a domestic surveillance platform covering roughly 25 million SIM cards. Additional incidents involved biological-risk grant writing, autonomous drone swarms, fabricated news streams, and large-scale fake-profile dating apps. Anthropic reported the bans, updated safeguards and passed threat indicators to industry partners and law-enforcement agencies.
Why it matters
The findings show AI can now automate sophisticated weapon design and mass surveillance, raising urgent security and ethical concerns.
In this story
