On June 9, 2026, Anthropic launched Claude Fable 5 — its most powerful AI model ever made available to the public. It topped nearly every benchmark, could play Pokémon from screenshots alone, and compressed months of engineering work into days. Three days later, it was gone.
At 5:21 PM ET on June 12, the US Commerce Department issued an export control directive ordering Anthropic to block all foreign nationals from accessing Fable 5 and its unrestricted sibling, Mythos 5. The scope was so sweeping — covering foreign nationals inside and outside the United States, including Anthropic’s own international employees — that the company had no choice but to disable both models for every user on the planet.
This is the first time in history that the US government has used export controls to recall a commercially deployed AI model. And it raises questions that will shape the future of artificial intelligence for years to come.
What Were Fable 5 and Mythos 5?
To understand the significance of the shutdown, you need to understand what got shut down.
Mythos 5 represents the ceiling of what Anthropic can build. First revealed through an accidental data leak in March 2026 and released as a preview to select partners in April, Mythos was described as possessing the strongest cybersecurity capabilities of any AI model in existence. It could autonomously discover and exploit software vulnerabilities at a level that, according to Anthropic’s own research, surpassed all but the most elite human hackers. The Guardian called it a potential “threat to global cybersecurity.” WIRED said it would “force a cybersecurity reckoning.”
Anthropic never intended Mythos for general release. It was too capable, too dangerous. Instead, the company restricted it to Project Glasswing — a collaboration with the US government where vetted cyber defenders used the model to patch critical infrastructure before bad actors could exploit it.
Fable 5 was the compromise. Built on the same underlying architecture as Mythos 5, it was designed to bring Mythos-class intelligence to the general public — with guardrails. When a user asked Fable 5 to do something sensitive (cybersecurity, biology, chemistry, or model distillation), the request was silently routed to Claude Opus 4.8, a significantly less capable model. Anthropic tuned these safeguards conservatively, admitting they would sometimes catch harmless requests — but the tradeoff, they argued, was worth it to get this level of intelligence into people’s hands safely.
And it was genuinely impressive. Stripe reported that Fable 5 compressed months of engineering into days on a 50-million-line Ruby codebase migration. It beat Pokémon FireRed using nothing but raw screenshots, no helper tools. It scored highest among frontier models on Cognition’s FrontierCode evaluation. At $10 per million input tokens and $50 per million output tokens, it cost less than half of what Mythos Preview had charged.
For 72 hours, it was the most capable AI model any civilian could use.
The Jailbreak That Wasn’t
The government’s justification centers on a jailbreak — a method of bypassing Fable 5’s safety guardrails. But the details matter, and they’re far less dramatic than the shutdown suggests.
According to Anthropic’s statement, the jailbreak in question is narrow and non-universal. It essentially involves asking the model to read a codebase and fix software flaws — the kind of thing thousands of developers do every day with existing AI tools. The vulnerabilities it uncovered were minor and previously known. Anthropic claims it verified that other publicly available models, including OpenAI’s GPT-5.5, can find the same issues without requiring any bypass at all.
“To date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak. We have not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result.”
Per reporting from Axios, a different company — not the government itself — claimed to have jailbroken Mythos, and the administration tried unsuccessfully to get Anthropic to pause Fable 5’s release before it ever went public. When Anthropic launched anyway on June 9, the government responded three days later with the nuclear option: export controls.
Anthropic’s Response: Compliance With Protest
Anthropic is complying with the directive. They have to — it carries the force of law. But they are not being quiet about their disagreement.
“We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users. However, 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. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
That last sentence is worth reading twice. Anthropic — the company that has built its entire brand on AI safety, that warned Mythos was too dangerous for the public, that instituted 30-day data retention policies specifically to catch jailbreaks — is now arguing that the government has gone too far.
The company also pointed out that the action contradicts principles it has publicly advocated for: that government intervention in AI deployments should be “transparent, fair, clear, and grounded in technical facts.”
The Deeper Irony
There’s a rich irony at play here that Forbes contributor Sandy Carter captured sharply: Anthropic spent months telling anyone who would listen that its models posed existential risks. The company’s April blog post on Mythos warned that “the fallout for economies, public safety, and national security could be severe” if such capabilities fell into the wrong hands. Fable 5 was the answer — safety through guardrails and monitoring.
Now the government has taken Anthropic at its word — and then some.
The contradiction cuts deeper. Earlier this year, the Pentagon designated Anthropic as a supply-chain risk, effectively barring it from Defense Department contracts. A federal judge suggested the designation looked retaliatory after Anthropic refused to let the military use Claude for mass surveillance and autonomous weapons. So here we have the same government that excluded Anthropic from defense work now invoking national security to lock its models away from the rest of the world. The company is simultaneously treated as a liability to do business with and a capability too potent to share.
What This Means for the AI Industry
The Fable 5 shutdown is not just an Anthropic story. It’s a precedent with industry-wide implications.
First, it demonstrates that the US government is willing to use export controls — a tool originally designed for physical goods like semiconductor equipment and weapons systems — to regulate software models in real time. The chips that power AI have been subject to export controls for years. Now the models themselves are fair game.
Second, it creates a chilling effect for enterprise AI adoption. Companies that built workflows around Fable 5 during its three-day lifespan had those workflows broken overnight. If frontier models can be recalled with a single Commerce Department directive, every enterprise AI deployment carries a new category of regulatory risk that didn’t exist a week ago.
Third, it complicates the international AI landscape. European observers are already asking whether similar measures could affect their access to American AI. If the US is willing to pull models from its own allies over narrow jailbreak concerns, what happens when geopolitical tensions are genuinely high?
Fourth, and most provocatively, it puts every AI company’s safety rhetoric under a microscope. Anthropic advocated loudly for regulation and got it in a form it now calls disproportionate. OpenAI, Google DeepMind, and others have made similar arguments about the need for government oversight. The Fable 5 precedent suggests that when governments do step in, the result may not look like the thoughtful, technically-informed process CEOs have described in congressional testimony.
What Happens Next
Anthropic says it is “working to restore access as soon as possible” and will share more details in the coming days. The company believes “this is a misunderstanding.”
But the damage, at least to the narrative of American AI leadership through open commercial deployment, is already done. A model that topped benchmarks across the board lasted 72 hours. The fastest shutdown in AI history is now also the most consequential — not for what the model could do, but for what its removal says about the emerging relationship between AI companies and the governments that regulate them.
The question isn’t really whether Fable 5 and Mythos 5 come back. It’s whether any frontier model can survive in a world where the government can pull the plug with a single letter — and whether that’s a world the AI industry, or the public, actually wants.
Sources: Anthropic, Anthropic Launch Blog, CNBC, Bloomberg, Fortune, Forbes, The Guardian, WIRED
