Anthropic's safety-first ethos collided with The Pentagon (scientificamerican.com)

From the article ...

Anthropic has drawn two red lines: no mass surveillance of Americans and no fully autonomous weapons ... OpenAI, Google and xAI—have agreed to loosen safeguards ... Pentagon has demanded that AI be available for “all lawful purposes.”

This seems to mostly be about the Pentagon sticking to the principle of not allowing a private corporation tell them what they can or can’t do.

Real tension here is governance, not model quality. If the Pentagon wants "all lawful purposes," vendors need contract-level guardrails: explicit prohibited uses, independent audit logs, and kill-switch rights. Otherwise "safety-first" is mostly branding.