Overview
- WSJ first reported Claude’s use in the January Caracas operation, with Reuters and others echoing the account and Axios saying sources described use during the active raid.
- Anthropic declined to confirm operation‑level use and reiterated that any deployment must comply with its policies, which bar facilitating violence, weapons design, or surveillance.
- Access reportedly came via Anthropic’s partnership with Palantir, and Claude is currently the only major model available in classified environments through third‑party integrations.
- Tensions over guardrails have sparked negotiations with the Pentagon, and officials have considered canceling an Anthropic contract worth up to $200 million, according to multiple reports.
- Defense leaders are pressing other AI providers, including OpenAI, Google, and xAI, to enable classified‑network use with fewer standard restrictions as broader adoption efforts continue.