Anthropic Says It Blocked Attempts to Use Claude for Bioweapons — and Caught Iran Using It to Surveil Its Own Citizens
In a single transparency report, Anthropic disclosed that it intercepted operations aiming to develop dangerous pathogens with Claude, and separately that Iran deployed the model to monitor thousands of its own people.
Anthropic published a transparency report this week that reads less like a corporate disclosure and more like an intelligence briefing. The company said it detected and blocked multiple operations attempting to use Claude to develop biological weapons, toxins, and — most alarmingly — to increase the contagiousness of existing viruses, according to @Ambitocom. The response was account suspensions and stronger filtering, but the more important signal is what the disclosure reveals about the threat surface of frontier models: people are actively trying, and the labs are catching some fraction of it.
The bioweapons angle would be the headline on most days. But the same reporting cycle surfaced a second, more concrete case. Anthropic revealed that Iran used Claude to help surveil thousands of Iranians, analyzing a corpus of 155,216 tweets and building a malicious Firefox extension, as documented by @ilex_ulmus. That post tied the tooling to the deaths of 156 Iranian civilians — a claim that should be treated with appropriate caution given the difficulty of establishing direct causal chains between a model's output and downstream real-world harm.
Get our free daily newsletter
Get this article free — plus the lead story every day — delivered to your inbox.
Want every article and the full archive? Upgrade anytime.
No spam. Unsubscribe anytime.