The White House Is Making Up Its Rules for AI in Real Time
TL;DR
The Trump administration forced Anthropic to take Claude Mythos and Claude Fable 5 offline through an export-control directive. Nearly a week later, nobody can point to the exact rule Anthropic allegedly broke. The reported triggers were concerns over SK Telecom’s access to Mythos and possible jailbreaks of Fable 5. Anthropic says it had coordinated the rollout with US officials and cut SK Telecom’s access immediately after objections.
Nauti's Take
The issue is not that the US government is watching powerful models closely. The issue is a rulebook being written while the fight is already underway.
If export control, national security, and jailbreak risk are serious concerns, labs need defined procedures, not back-channel pressure and retroactive shutdowns. Otherwise safety becomes a black box and every model launch turns into a political stress test.
Briefingshow
The episode shows how quickly anti-regulation rhetoric can turn into informal control over AI launches. If companies cannot tell which line they crossed, they will optimize less for clear safety compliance and more for political signaling. For OpenAI, Google, Meta, and other labs, the new rule may be simple: brief Washington early or risk being frozen out after launch.
Sources