Anthropic's Claude Mythos and Fable 5 have now been offline for nearly a week, and the Trump administration still has not publicly stated which rule the company broke. The export control directive issued on June 18, 2026 forced one of the leading US frontier labs to pull its most advanced models, and after days of negotiation between Anthropic and the White House, the two sides remain at odds over the terms for bringing the models back. The only on-the-record signal from the administration is a post on X from White House technology adviser David Sacks outlining the general situation.
Anthropic does not believe it violated any concrete procedure laid out by the Trump administration, according to a person close to the company. The White House position is that Anthropic behaved recklessly and cannot be trusted to safely roll out frontier technology. Neither account survives scrutiny without the other, because the procedures themselves were never written down. This is the core problem: there is no rulebook, only the directive.
The administration has spent the year rolling back AI guardrails on the argument that they would hand the lead to China. President Trump signed executive orders reversing the Biden-era national AI framework and created a federal task force to challenge state AI laws deemed onerous. The vacuum left behind has now produced its first enforcement action — improvised in real time against the lab that has marketed itself the hardest on safety.
Key facts
- 01Claude Mythos and Fable 5 have been offline for nearly a week after a June 18, 2026 export control directive from the Trump administration.
- 02The directive bars all foreign nationals — including many of Anthropic's own employees — from accessing the two frontier models.
- 03Customers locked out include Apple, Meta, and much of the Fortune 500.
- 04Concerns trace to Anthropic sharing Mythos with SK Telecom and Amazon CEO Andy Jassy flagging jailbreak risks on Fable 5 to Treasury Secretary Scott Bessent.
- 05Trump signed an executive order last month creating a 'voluntary' early-testing system with an explicit carve-out against a licensing regime.
The trigger appears to be two separate concerns. Earlier this month, US officials learned Anthropic had shared Mythos with SK Telecom, the South Korean telecom giant officials allege has ties to China. Separately, Amazon CEO Andy Jassy raised concerns to Treasury Secretary Scott Bessent that guardrails on Claude Fable 5, the safeguarded variant of Mythos, could be circumvented. Anthropic says it coordinated with the US government on the Mythos rollout and has worked with SK Telecom for years without incident; when the White House flagged the relationship, Anthropic revoked access immediately.
The collateral damage is substantial. The directive prohibits all foreign nationals from accessing Mythos and Fable 5, which locks out a meaningful share of Anthropic's own staff from the models the company says have accelerated internal research and development in recent months. Every external customer is locked out as well, including Apple, Meta, and much of the Fortune 500. An anti-regulatory administration has, in effect, taken two frontier products off the US market by phone call.
Executives at OpenAI, Google, Meta, and Cohere are now triangulating from Anthropic's misfortune. The consensus emerging from those conversations is that labs will need to give the White House early access to upcoming models and proactively brief officials before any launch, on the theory that surprising the administration is the unforgivable error.
That posture is exactly what the Trump administration said it would not require. Last month's executive order created a voluntary system for labs to submit models for early government testing, with an explicit carve-out stating it would not become a mandatory licensing regime — a carve-out the tech industry insisted on before the order could move. The Anthropic episode has now produced the regime the carve-out was supposed to prevent, just without the rules, timelines, or appeal process that an actual licensing system would include.
On the technical merits, the jailbreak concern is unlikely to have a clean resolution. Large language models are probabilistic systems, and independent cybersecurity researchers have repeatedly shown that every frontier model can be jailbroken to some degree. If Fable 5 has to wait for jailbreaks to be solved before it can ship, it will not ship.
The standoff matters beyond Anthropic because it sets the operating template for every other US lab. An ad hoc licensing regime with no published criteria gives the executive branch enormous discretionary power over which models reach the market and when — power that can be applied evenly or unevenly, predictably or not, depending entirely on who is in the room. For frontier labs, the rational response is to over-disclose, slow launches, and route product decisions through Washington. That is a cost American AI companies will now carry whether or not Mythos and Fable 5 come back online, and it is a cost their Chinese competitors will not.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
