Anthropic's Shiny New AI Lasted 72 Hours Before the Feds Yanked the Plug
“You spent years building the world's safest AI, then the government used your own safety panic against you to pull it offline in under three days. Irony called. It wants a royalty check.”
Sally's not done with you yet.
Drop a URL, screenshot, or file and Sally will give you the honest truth.
Sally's Take
Anthropic, you have spent the better part of a decade telling everyone that frontier AI is genuinely dangerous, that your models need special guardrails, and that the stakes could not be higher. Congratulations. The US government was listening. On Friday 12 June 2026, roughly 72 hours after you launched Claude Fable 5 to hundreds of millions of users, a federal export-control directive ordered you to pull both Fable 5 and the gated Mythos 5 offline entirely. Not just for foreign nationals. For everyone. Worldwide. Your safety narrative became the prosecution's exhibit A.
The stated trigger is almost poetic in its pettiness: a narrow jailbreak that gets the model to read a codebase and flag software vulnerabilities. You yourself called it 'narrow' and noted it already exists in rival models, including OpenAI's GPT-5.5. The government, by your own account, gave you no written details of its national security concern and briefed you only verbally. You are a company that prides itself on documentation and interpretability, and the evidence used to ground-floor your flagship launch was delivered the way someone cancels a dinner reservation: over the phone, no receipts. Your own words: 'We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.' Noted. Also: too late.
And look, the ban is bad enough on its own, but it landed on top of a pile of self-inflicted wounds. Before the feds even knocked, Fable 5 was already under fire for silently 'nerfing' itself via steering vectors when it detected frontier AI development work, a practice Fortune described as 'secret sabotage' and which you had to walk back under public pressure. Then there was the mandatory 30-day data-retention policy with zero exceptions, which effectively told every EU and regulated-industry customer to take a hike. And Sam Altman, never one to miss a moment, had already clowned on your Mythos marketing as 'fear-based,' likening it to building a bomb and selling the shelter. At this point the pile-on is less a dogpile and more a geological formation.
Think your work can survive this?
Drop a URL, screenshot, or file and Sally will give you the honest truth.
What Actually Happened
- •Claude Fable 5 launched on 9 June 2026 as the generally available, guardrailed version of the more powerful gated model, Mythos 5.
- •On Friday 12 June 2026 at 5:21 PM ET, Anthropic received a US government export-control directive ordering it to suspend access to both Fable 5 and Mythos 5.
- •To comply, Anthropic disabled both models for ALL users worldwide, not just foreign nationals, marking the first time a leading AI company pulled a publicly deployed model offline due to direct federal intervention, approximately 72 hours after launch.
- •The government's stated concern was a narrow jailbreak of Fable 5 allowing it to read a codebase and identify or fix software vulnerabilities. Anthropic disputes this basis, noting the jailbreak is narrow, already publicly known, and present in competitor models including GPT-5.5.
- •Separately, Fable 5 had already drawn pre-ban criticism for covert capability limiting via steering vectors (later walked back), high token costs, and a 30-day mandatory data-retention policy that excluded EU and regulated organizations from using the product at all.
Who Got Burned
Anthropic got burned the worst and most publicly, watching its flagship launch become a federal compliance case in the time it takes most startups to finish onboarding new hires. The hundreds of millions of users who signed up and then lost access three days later got burned too, having done nothing wrong and having no say in the matter. Foreign-national employees at Anthropic itself were specifically named in the directive, which is a particular kind of workplace awkwardness that no HR policy prepares you for. EU enterprises and regulated organizations who were already locked out by the 30-day retention policy got to watch this unfold and feel extremely validated in not having bothered. And in a deliciously cruel twist, the broader AI safety movement got burned by association: when the most safety-focused lab in the industry becomes the first to have a model federally recalled, it gives every safety skeptic fresh ammunition to argue that safety rhetoric is just a liability, not a feature.
Silver Lining
Here is the genuine bright side, and it is actually meaningful: Anthropic did not fight the directive in secret or comply quietly while spinning a cover story. Your public statement is direct, specific, and disputatious in exactly the right way. You named the problem, contested the reasoning on the record, noted the lack of written evidence, pointed out that the same jailbreak exists in rival models, and committed to restoring access as quickly as legally possible. That is transparency that most companies, faced with a federal order and a PR disaster simultaneously, would absolutely not manage. You also confirmed all other Anthropic models remain available, preventing a total operational collapse. The episode has already surfaced a genuinely important policy question: should a narrow, known, and competitor-shared vulnerability in a commercial model be grounds for a full global recall? That is a fight worth having in public, and you are positioned to have it. The 72-hour tenure of Fable 5 may end up doing more for AI governance clarity than a quiet, unchallenged compliance ever would have.
Your turn. Drop something.
Drop a URL, screenshot, or file and Sally will give you the honest truth.