Anthropic Says Iran and Houthi-Linked Users Tried to Weaponize Claude

A new Anthropic security disclosure details blocked accounts attempting to develop guided weapons and run influence campaigns targeting U.S. Navy bases — the clearest evidence yet that adversarial states are actively probing Western models.

Anthropic has disclosed that it blocked accounts tied to Iranian influence operations and users in northern Yemen who attempted to use Claude's coding tools to develop advanced guided weapons, according to circulating summaries of the company's latest security report. The details are striking not because state actors want Western AI — that was always going to happen — but because of how specifically they were caught trying to use it.

Unlock the full briefing

Get every story in today's briefing, the full archive, and the daily AI intelligence brief.

All stories today

Full archive

Daily brief

Cancel anytime. Payments powered by Stripe.