The White House Is Making Up Its Rules for AI in Real Time
TL;DR
Anthropic still cannot distribute Claude Mythos or Claude Fable 5 after the Trump administration sent an export-control directive. The unresolved issue: no clear public rule explains what Anthropic allegedly violated. US officials are reportedly concerned about Mythos access for SK Telecom and possible jailbreaks in Fable 5. Anthropic says it revoked SK Telecom’s access immediately after officials raised concerns.
Nauti's Take
This is not stable AI regulation; it is policy by emergency handling. The safety concerns may be real, especially around export controls and jailbreaks.
But if no one can clearly state which line was crossed, the incentive gets messy: labs perform for Washington early instead of meeting clear technical and legal gates. For users and enterprises, the risk is no longer just model behavior; it is supply-chain uncertainty.
Briefingshow
The case shows how quickly voluntary AI safety reviews can turn into a de facto approval system. If officials intervene without clear, published criteria, labs face a planning problem: they are no longer optimizing only for safety and product readiness, but also for political signals that can shift in real time.
Sources