DooDooLamb News
The White House Is Making Up Its Rules for AI in Real Time
Brief published June 20, 2026 ยท Original source published June 18, 2026
Original reporting by Maxwell Zeff at wired.com.
Automated brief. Verify important details at the original source.
The White House Has AI Export Rules, Which Is Impressive Because It Also Has No AI Export Rules
The Trump administration sent Anthropic a strongly worded letter about its AI models, and somehow the most advanced AI company in the world cannot get a chatbot to explain what it actually says.
It has been nearly a week since the Trump administration dropped an export control directive on Anthropic like a parking ticket left under the windshield wiper of a Formula 1 car. Claude Mythos and Fable 5, Anthropic's most capable models, remain offline for distribution. Nobody at Anthropic knows exactly what they did wrong. Nobody at the White House is in a rush to clarify. And the rest of the AI industry is watching with the specific frozen-smile energy of someone who just watched a coworker get called into HR and is very carefully not making eye contact with anyone.
The administration's directive invoked export control authority, which is the government's mechanism for preventing sensitive technology from ending up in the hands of adversaries, foreign militaries, or anyone else on a list that gets updated roughly as often as your DMV's website. Export controls are serious business; they've been used to restrict semiconductor exports to China and to keep certain weapons technologies from proliferating. Applying them to AI models is newer territory, and the rules there are, to use the technical regulatory term, vibes. The Trump administration has been assembling this particular airplane while also arguing about whether planes should exist, and Anthropic just got sucked into an engine.
What makes this particularly surreal is that Anthropic cannot get a straight answer about the specific violation. According to reporting from Wired, the company received a directive, pulled its models, and then entered the bureaucratic equivalent of being put on hold by a phone tree that doesn't have a department for your problem. This is not a small company being crushed by paperwork it doesn't understand. Anthropic has serious policy staff, former government officials, and the kind of Washington infrastructure you build when you know regulation is coming and you've decided to run toward it rather than away from it. The company has spent considerable energy positioning itself as the responsible actor in AI development, the one that talks to senators and publishes safety research and generally tries to be the AI lab your mom would trust. And it still can't get someone on the phone to explain the charge.
The models themselves deserve a moment of context. Claude Mythos and Fable 5 aren't chatbots in the sense that most people understand chatbots. These are frontier models, which is the industry's way of saying they represent the current outer boundary of what AI systems can do. Restricting their distribution isn't like pulling a mobile app from an app store. It's closer to telling a pharmaceutical company it can't ship a drug while also declining to specify which regulation it violated. The practical consequences ripple outward fast: developers who've built products on these models are stuck, enterprise customers are fielding awkward questions, and Anthropic is burning runway in the form of opportunity cost while waiting for a bureaucrat to return a call.
What nobody seems willing to say out loud, but what the situation makes fairly obvious, is that the administration is constructing its AI governance framework in approximately real time. The executive orders, the export directives, the Commerce Department guidelines, all of it is being written by people who are simultaneously learning what these systems actually do. That's not necessarily a condemnation; novel technology genuinely is hard to regulate before you understand it. But there's a meaningful difference between "we're carefully developing rules as we learn" and "we sent a directive and we'll figure out what it means later," and the Anthropic situation suggests the latter. The company is being held to standards that have not been fully articulated, which is the regulatory equivalent of getting a red card in a sport where the referee is still reading the rulebook.
The reality check here is uncomfortable for everyone involved. For Anthropic, the lesson is that being the responsible AI lab, the one that cooperates with Washington and voluntarily limits its own capabilities and publishes safety cards, does not actually buy you clarity when the government decides to act. It might buy you goodwill eventually, but goodwill doesn't get Claude Mythos back online. For the broader industry, this is a preview of what AI regulation looks like when it's improvised rather than designed: companies operating under rules that shift without notice, making product decisions based on their best guess about what the government currently wants, and hoping their lawyers are reading the right tea leaves. For developers and founders who've built on these models, the message is that the infrastructure you depend on can disappear for reasons that nobody will fully explain to you, which is a fun thing to discover when you have a product demo next Tuesday.
The funniest and most depressing part of all this is that Anthropic named its models Mythos and Fable, words that mean, roughly, stories that aren't true. Whether that's irony or prophecy depends on whether you think the company's safety-first positioning will ultimately save it from exactly this kind of arbitrary regulatory whiplash. Right now the answer seems to be no, but the story isn't over, and in Washington, the sequel is always worse than the original.