Claude Military Use Exposes Policy Split

The contradiction is mostly surface-level: Anthropic ships one model family across multiple deployment surfaces, so consumer refusals don't map cleanly to government contracts. Once you separate claude.ai from enterprise and public-sector deployments, the mystery gets smaller.

• –Anthropic says its tailored government exceptions still exclude weapons design, domestic surveillance, censorship, and malicious cyber operations.
• –Claude Gov and other government-facing surfaces show the same brand can run under very different guardrails.
• –Recent reporting on Claude in the Iran/Middle East context shows how quickly AI becomes operational infrastructure when the customer is a defense agency rather than a consumer.
• –For developers, the lesson is to read deployment terms, not just demo behavior, before assuming what a model can or cannot do.