Anthropic’s Mythos mess is only getting worse
TL;DR
Anthropic can bring Mythos 5 back after two weeks of talks, but only for a small group of approved organizations. Fable 5, the more public Mythos-class model, is still stuck with no clear timeline for broader release. The June 12 export-control order remains in place. The Trump administration granted an exception rather than fully reversing course. The case shows a messy new US playbook: frontier AI models are not banned outright, but access can be negotiated politically and approved customer by customer.
Nauti's Take
The Mythos fight looks like regulation by panic button. Scrutinizing powerful cyber models is legitimate, but a half-blocked, half-approved exception system is not a durable framework.
Anthropic gets part of Mythos 5 back, yet the real damage is the precedent: teams can train a frontier model and still not know whether launch depends on engineering, business readiness, or political clearance.
Briefingshow
This is bigger than Anthropic. If frontier models need case-by-case political clearance, AI labs lose predictability, customers lose access, and security teams lose tools they may actually need. It also sends a global signal that US AI access can be interrupted by Washington with little warning.
Sources