Overview
- In sworn declarations, Anthropic’s public‑sector lead said the company has no remote kill switch, backdoor, or access to alter Claude once deployed in government systems, and cannot view user prompts or push updates without approval.
- Anthropic’s policy chief said the firm never sought approval power over military operations and produced a March 4 email from a Pentagon undersecretary saying the sides were “very close” on surveillance and autonomy issues.
- Government filings maintain the supply‑chain risk designation, arguing Anthropic could preemptively alter or disable its model during operations and noting steps with cloud providers to block any unilateral changes.
- A new Pentagon declaration flags Anthropic’s employment of foreign nationals, including workers from China, as an adversarial risk, while Anthropic points to cleared personnel and prior security measures in classified environments.
- Major tech companies and worker groups have filed or signaled support for Anthropic, warning of broader industry impacts, as the San Francisco court prepares to hear the company’s request for temporary relief.