Story perspectives

Anthropic blocks Houthi-linked missile-guidance projects using Claude

By Drooid · Sep 11, 2026, 10:03 AM UTC · How we work

Summary

Story summary

Anthropic blocked a Houthi-linked group that used Claude to develop missile-guidance software, running three projects Dec 2025-Aug 2026.
Anthropic barred Russian APT29 operators who employed Claude for cyber-espionage targeting Ukraine, Europe and diplomatic entities.
Anthropic banned Chinese-aligned accounts that illicitly distilled Claude to train Alibaba models, involving over 151 million exchanges.
Despite Anthropic’s challenge to a Pentagon blacklist, the DoD continues using Claude in missions over Iran and Venezuela.

History

Historical Context

In 2010 the United States, led by General James Cartwright at US Strategic Command, and Israel’s Unit 8200 launched Operation Olympic Games, deploying the Stuxnet worm that stopped about 1,000 of Natanz’s 5,000 centrifuges. The leak of Stuxnet sparked international policy debates on cyber weapons, a backdrop for AI-driven defenses such as Anthropic’s 2025-2026 blocks against state-linked threat actors targeting Ukraine, Europe and the Pentagon.

1 of 2

Story summary

Anthropic blocked a Houthi-linked group that used Claude to develop missile-guidance software, running three projects Dec 2025-Aug 2026.
Anthropic barred Russian APT29 operators who employed Claude for cyber-espionage targeting Ukraine, Europe and diplomatic entities.
Anthropic banned Chinese-aligned accounts that illicitly distilled Claude to train Alibaba models, involving over 151 million exchanges.
Despite Anthropic’s challenge to a Pentagon blacklist, the DoD continues using Claude in missions over Iran and Venezuela.

1 / 2