too powerful to release, then released anyway
In a rare and telling moment for the AI industry, Anthropic was ordered by US federal authorities to suspend its most powerful model, Claude Fable 5, just days after its public release — the very model the company had once deemed too dangerous to release at all. A discovered method for bypassing its safety guardrails prompted the withdrawal, though Anthropic maintains the vulnerabilities are unremarkable and common across the industry. The episode unfolds against a backdrop of deepening institutional conflict, as the Trump administration has already branded Anthropic a 'supply chain risk' — a designation without precedent for an American technology firm — raising enduring questions about who holds the authority to define danger in the age of artificial intelligence.
- A jailbreaking technique capable of bypassing Fable 5's safety guardrails was discovered and handed to federal authorities within days of the model's public launch.
- The suspension is especially jarring because Anthropic had spent months warning the world that Fable 5 was uniquely powerful — even refusing to release it at first — only to have those warnings turned against the company by regulators.
- Anthropic is pushing back, arguing the exploited vulnerabilities are minor and no worse than flaws found in competing models already available to the public.
- The forced withdrawal affects every user worldwide, not just government customers, marking the first time a major AI company has been compelled to pull a product from the market over security concerns.
- The suspension deepens an already bitter standoff: the Trump administration has labeled Anthropic a 'supply chain risk,' the Pentagon tried to bar government use of its tools, and the company is now fighting both battles simultaneously in court and in public.
Anthropic pulled Claude Fable 5 from public access this week, just days after its launch, following a federal order tied to a discovered jailbreaking vulnerability. The timing was striking: Fable 5 was the model Anthropic had spent months warning the world about — so capable, so potentially dangerous, that the company had initially refused to release it at all. When it finally did, the safeguards it had so publicly championed became the very thing under scrutiny.
Federal authorities presented Anthropic with a technique for bypassing Fable 5's security restrictions, and the company responded by disabling both Fable 5 and its related model, Mythos 5, for all users globally. Anthropic reviewed the method and concluded the vulnerabilities were minor — the kind found in other publicly available AI systems, exploitable without any special bypass. The company's position was that the government's alarm was disproportionate to the actual risk.
But the suspension cannot be separated from a broader political conflict. The Trump administration has been openly hostile toward Anthropic for months. Defence Secretary Pete Hegseth designated the company a 'supply chain risk' — a label historically reserved for foreign adversaries, never before applied to an American tech firm — effectively barring government agencies from its tools. Anthropic sued the Pentagon, and a federal judge allowed government use of its models to continue while the case proceeds.
Now, with the jailbreaking order, Anthropic faces pressure that reaches far beyond Washington. The forced withdrawal of Fable 5 sets a precedent: a major AI company compelled to remove a product from the market on security grounds. The question it leaves behind — who has the authority to decide when an AI model is too dangerous, and what that judgment means for the companies building them — is one the industry will be answering for years to come.
Anthropic pulled the plug on Claude Fable 5 this week, just days after making it available to the public. The company had spent months building anticipation for what it called its most powerful AI model yet—so capable, so potentially dangerous, that it initially refused to release it at all. Then, abruptly, US authorities ordered the suspension. The reason: someone had found a way around the safety guardrails.
The company announced the decision on its website with a terse explanation. Federal authorities had discovered a method for "jailbreaking" Fable 5, a technique that bypasses the software restrictions meant to keep hackers from accessing sensitive systems or unlocking restricted features. In response, Anthropic said it had no choice but to disable both Fable 5 and its related model, Mythos 5, for all users worldwide to stay in compliance with the order.
The irony was sharp. Anthropic had spent the lead-up to release emphasizing the safeguards it had built into the system. The company had even conducted a limited preview in April, inviting a small group of organizations to test the model and hunt for vulnerabilities before the public launch. Finance leaders, tech executives, and government officials had all weighed in with concerns about letting such a powerful tool loose. Anthropic's own marketing had leaned hard into the danger: this was a model "too powerful to release," the company had said, a system whose capabilities exceeded anything it had ever made available to the general public.
When authorities showed Anthropic the jailbreaking technique, the company reviewed it and concluded the vulnerabilities being exploited were minor—the kind of flaws that other publicly available AI models also contained. Anthropic argued that discovering these weaknesses didn't require any special bypass method; competitors' systems could find them just as easily. The company's position was essentially that the alarm being raised was overblown, that it had done what any responsible developer would do, and that the government's response was disproportionate.
But the suspension sits within a much larger conflict. Anthropic has been in the Trump administration's crosshairs for months. The president has criticized the company publicly. Pete Hegseth, the US Defence Secretary, went further, labeling Anthropic a "supply chain risk"—a designation historically reserved for foreign companies in adversarial nations, never before applied to a US technology firm. That label effectively barred government agencies from using Anthropic's tools.
Anthropologic is fighting back. The company sued the Pentagon over the supply chain designation, and a federal judge ruled that the Pentagon's directive could not be enforced while the lawsuit proceeds. That means, technically, government agencies can still use Anthropic's models. But now, with the jailbreaking order, the company faces a different kind of pressure: one that affects not just government customers but every user worldwide.
The suspension of Fable 5 marks a rare moment when a major AI company has been forced to withdraw a product from the market due to security concerns. It raises a question that will likely define the next phase of AI regulation: who decides when a model is too dangerous to use, and what does that decision actually mean for the companies building these systems and the people relying on them?
Citas Notables
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance— Anthropic statement
Our understanding is that the government believes it has become aware of a method of bypassing, or 'jailbreaking' Fable 5— Anthropic statement
La Conversación del Hearth Otra perspectiva de la historia
Why did Anthropic release something it had just called too dangerous to release?
That's the tension at the heart of this. The company built real safeguards and tested them with a small group first. They believed they'd done the work. But there's always a gap between what you test in private and what happens when millions of people have access.
And the jailbreak they found—was it actually serious?
That's where it gets murky. Anthropic says the vulnerabilities were minor, things other models have too. But the government saw it differently. Maybe they were right to be cautious. Or maybe they were looking for a reason to tighten the screws on a company they already didn't trust.
The "supply chain risk" label—that's extraordinary, isn't it?
Completely unprecedented for a US company. It's the kind of thing you'd use against a Chinese firm or a Russian one. Applying it to Anthropic signals something deeper: the administration views this company as a threat to national security, not just a business rival.
So is this about the jailbreak, or is it about politics?
Probably both. The jailbreak gave them legal cover to act on something they already wanted to do. But the jailbreak itself is real—it exists, it works. The question is whether it's a reason to suspend the entire product or a reason to patch it and move on.
What happens to Anthropic now?
They keep fighting in court. They keep developing. But they've lost momentum, lost trust, lost access to a huge market. And every other AI company is watching to see if this becomes the template for how the government handles models it doesn't like.