Overview
- Anthropic published a 154‑page threat report on Sept. 14 that says it found Iran‑nexus actors using Claude to profile hundreds of Israelis and members of the Jewish diaspora and to automate influence operations.
- The report details cases in which Iran‑aligned actors used Claude for weapons guidance and naval vulnerability research and says the company blocked biological‑related queries while withholding specific institution names.
- Anthropic says it banned the implicated accounts, built new detection tools, and shared its intelligence with government authorities to disrupt the campaigns.
- Enterprises such as Palantir, Nvidia and Booz Allen have limited staff use of Anthropic and OpenAI models and demanded irrevocable zero‑data‑retention guarantees, prompting Anthropic to offer Enterprise Frontier Safeguards that let customers store activity logs under their own keys.
- The disclosure has sharpened industry debate over coordinated pacing, third‑party model testing and a possible standards body, and it is shifting customers toward private or air‑gapped solutions while Anthropic continues IPO preparations for late 2026.