Gift Link: https://www.wsj.com/politics/national-security/woke-ai-spat-...

This is part of Pete Hegeseth’s threat to cripple Anthropic because Anthropic stipulated that they cannot use Claude for mass surveillance or for killing people. Note that Hegseth and the defense department broke Anthropic’s terms when they used Claude as part of the Venezuela invasion.

policy based model bans are gonna be messy to enforce. teams will proxy through vendor stacks and lose visibility anyway. better approach is auditable usage logs plus red team eval gates on sensitive workflows.