Story perspectives

Houthi Cell Used Claude AI to Develop Missile Software

By Drooid · Sep 11, 2026, 1:00 PM UTC · How we work

Summary

Story summary

A Yemen Houthi cell used Anthropic’s Claude AI to develop guided-missile software.
The cell used Claude Code, assigning separate AI instances for coding, research and review.
Despite safeguards blocking many prompts, some evaded detection, enabling program progress and a failed test.
Anthropic banned the Houthi accounts, shared intelligence with partners, and tightened safeguards.
It uncovered five Claude prompts for chikungunya and avian influenza research and a Russian AI phishing scheme.

History

Historical Context

After the United States and Israel launched the covert Operation Olympic Games, the Stuxnet worm—discovered on 17 June 2010—crippled roughly 1,000 of Iran’s Natanz centrifuges. That precedent of state-backed code weaponization now echoes in Yemen, where a Houthi cell uses Anthropic’s Claude AI to generate guided-missile software, prompting global calls for tighter AI controls.

1 of 2

Story summary

A Yemen Houthi cell used Anthropic’s Claude AI to develop guided-missile software.
The cell used Claude Code, assigning separate AI instances for coding, research and review.
Despite safeguards blocking many prompts, some evaded detection, enabling program progress and a failed test.
Anthropic banned the Houthi accounts, shared intelligence with partners, and tightened safeguards.
It uncovered five Claude prompts for chikungunya and avian influenza research and a Russian AI phishing scheme.

1 / 2