The Anthropic export control order landed on a Friday evening: at 5:21 p.m. ET on 12 June 2026, the Commerce Department’s Bureau of Industry and Security, under the signature of Commerce Secretary Howard Lutnick, directed Anthropic to suspend all access to its two most advanced models, Fable 5 and Mythos 5, for any foreign national anywhere in the world, including Anthropic’s own foreign-national staff.
Anthropic’s response was blunt: without a way to verify every user’s nationality in real time, the only workable option was to pull both models entirely.
When Amazon Made the Call That Started It All
The chain of events that produced the order traces back to the day before. On 11 June, Amazon CEO Andy Jassy raised concerns about security risks in Anthropic’s models during a pre-scheduled call with senior Trump administration officials that had been arranged to discuss an unrelated topic, according to sources familiar with the matter. Amazon researchers had reportedly found a method to bypass Fable 5’s guardrails.
Fable was designed as a public-facing version of Mythos, equipped with strict guardrails intended to block use in biology, chemistry, and cybersecurity, and to prevent third parties from distilling the model to re-create it. The Amazon research paper underpinning those concerns has not been made public.
What makes the sequence jarring, according to a source familiar with Anthropic cited by Fortune, is that the government’s warning arrived with no prior communication or indication that it considered Fable a national security threat. From Anthropic’s perspective, the order came out of nowhere.
The Anthropic Export Control Order and the Question of Who Gains
Cybersecurity professionals were fast to push back. Dozens of cybersecurity experts signed an open letter calling on the Trump administration to revoke the order, arguing that removing advanced AI capabilities from network defenders in the United States creates its own security risk. Katie Moussouris, chief executive of Luta Security and a signatory to the letter, explained that the jailbreak method Amazon’s researchers demonstrated involved asking the kinds of questions that normal defenders would routinely put to an AI model: behaviour the model was explicitly built to handle.
Anthropic itself noted that similar jailbreaks had been identified in other AI models, sharpening the question of whether this was ever a straightforwardly technical judgement.
On Capitol Hill, the order drew pointed responses. Virginia Senator Mark Warner, the senior Democrat on the Senate Intelligence Committee, highlighted the administration’s quarrel with Anthropic. Representative Thompson went further, arguing that the order adds to evidence that the Trump administration lacks a coherent plan for managing the cybersecurity risks of frontier AI.
The political dimension is hard to separate from the technical one. As Sean O’Kane observed on TechCrunch’s Equity podcast, Anthropic ‘has not had the best relationship with the Trump administration in a way that stands apart from the other leading AI labs.’ A government lawsuit against the company, a prior designation as a supply chain risk, and now a Friday-evening export order: the pattern sits uncomfortably alongside the administration’s professed ambition to keep the United States ahead in AI.
For Anthropic’s competitors, the picture is mixed. The more charitable reading is that a company prepared to avoid antagonising the White House can operate with fewer interruptions. The less comfortable reading is that the regulatory landscape now depends, at least in part, on whether the right people return calls. That is not a durable framework for an industry building products it describes as transformative.
A Trump administration official, cited by Axios and flagged in subsequent coverage, suggested the intent may be broader than one company: the position, as reported, is that any model at the capability level of Mythos or above would need to clear the administration’s national security apparatus before deployment. If that framing holds, the Anthropic order could be less a targeted reprisal and more the opening move of a policy for regulating frontier models generally.
There is one further irony Anthropic’s team will be reluctant to acknowledge publicly. A previous clash between the company and the administration was followed by a measurable uptick in downloads of its Claude assistant, as users drawn to the controversy looked for an alternative to ChatGPT. Being cast as the dangerous one, the model so powerful that even the government moved to contain it, is not straightforwardly bad for a consumer brand. Rebecca Bellan put it plainly on the podcast: ‘Everyone loves a bad boy.’
The models remain offline. Whether they return in modified form, whether the order is revised after cybersecurity experts’ lobbying, or whether the administration extends Mythos-level scrutiny to other labs will determine whether this episode is remembered as a political dispute or the moment AI regulation in the United States took a structural turn.
