Claude Gov: a model built for the enclave
In June 2025 Anthropic announced Claude Gov, a set of models built for US national security customers. The company says they are already deployed by agencies at the highest level, and that access is limited to those working in classified environments.
Why it matters
The line moved again: from whether a model may be sold to the military, and what it may do inside an enclave, to a separate line of models that does not exist outside one. This is the first record in the atlas of a model whose capabilities cannot be independently checked at all.
Anthropic's wording on access: "access to these models is limited to those who operate in such classified environments". By the same text the models are "already deployed by agencies at the highest level of U.S. national security". What they were tuned for: strategic planning and operational support, intelligence analysis and threat assessment, handling classified material with fewer refusals, understanding intelligence and defence documents, languages and dialects that matter for national security, and interpreting complex cybersecurity data. Anthropic states the models underwent the same safety testing as the rest of the Claude line. That claim cannot be checked from outside, and the page does not say what these models may not be used for. Both facts are recorded here as facts about the record, not as a reproach.