Anthropic’s latest model access change came with a new security requirement after government restrictions on Fable 5 and Mythos 5 were removed. The condition shows that frontier model releases are increasingly negotiated around concrete controls, not only published safety claims.
For AI teams, the important signal is that deployment policy can change when providers add technical mitigations. Safety classifiers, monitoring, and abuse controls are becoming part of the release process for high-capability models.