David Sacks, co-chair of the President’s Council of Advisers on Science and Technology, has gone public with the Trump administration’s account of what led to Friday’s emergency export controls on Anthropic’s Fable 5 and Mythos 5 models — and his version puts Dario Amodei directly at the center of the decision.

In a detailed post on X, Sacks laid out a sequence of events starting with a trusted partner of both Anthropic and the US government coming forward with a working jailbreak of Fable’s guardrails during testing. According to Sacks, the administration asked Amodei to either fix the vulnerability or pull the model from deployment. Amodei refused. The export control followed.

“The Admin asked Dario to fix the jailbreak or de-deploy the model,” Sacks wrote. “Dario refused.”

What Fable Is, And Why The Guardrails Matter

Fable 5 is essentially Mythos with guardrails — a consumer-facing version of Anthropic’s most capable and most restricted model, stripped of the cyber capabilities that made Mythos too sensitive for public release. The guardrails are what separate a product you can deploy broadly from one that Anthropic itself argued needed regulatory treatment similar to a weapon.

That framing matters here. Sacks’ position is that if those guardrails fail, you haven’t just shipped a flawed product — you’ve exposed Mythos and its advanced cyber capabilities to whoever can find the vulnerability. And Anthropic, having made the case that Mythos was dangerous enough to warrant regulation and guardrails in the first place, is responsible for making sure those guardrails actually hold.

Anthropic’s Counter

Anthropic has pushed back on Sacks’ characterization. The company says it reviewed a demonstration of the reported jailbreak and found it produced only minor, previously known vulnerabilities — ones that other publicly available models could already surface without any bypass at all. Its position is that the technique amounted to pointing the model at a codebase and asking it to identify and fix security flaws, which it describes as a narrow and non-universal jailbreak.

The company has said it only received verbal evidence from the government, with no documentation or written specifics provided before the directive landed. According to reporting from DNYUZ, administration officials called Anthropic at 1:15 p.m. on Friday, gave them 90 minutes to respond, and then at 5:21 p.m. notified the company that export controls were already being imposed. Anthropic complied and took both models offline globally — affecting all customers, including those in markets like India where Claude had been growing rapidly — to ensure it could meet the terms of the directive.

Sacks finds the company’s framing difficult to reconcile with its public track record: “That is not what the trusted partner and the USG believe; nor is that kind of minimizing language consistent with Anthropic’s brand as the AI safety company.”

The Contradiction Sacks Is Pointing At

The sharper edge of Sacks’ post is the consistency argument. Anthropic spent years building an identity around taking safety more seriously than anyone else in the industry. Dario Amodei has said publicly that he sees a 25% chance things go “really, really badly” with AI. Anthropic championed export controls, advocated for government regulation of frontier models, and positioned Mythos as something dangerous enough that only a small circle of trusted partners should ever have it.

Now, when a government partner flags a jailbreak and asks them to fix it or pull the model, Anthropic’s response is that the issue isn’t serious enough to act on. Sacks is saying you can’t have it both ways — either the guardrails matter or they don’t.

“In the past, Anthropic has always said that safety must be top priority and taken super seriously. In this case, Anthropic prioritized the continued offering of the consumer model over safety,” he wrote.

The export control, in Sacks’ telling, was a last resort. He says the administration was reluctant to act and remains genuinely puzzled that Anthropic hasn’t moved to cooperate, given that a fix would presumably get the model back into general release — which is what both sides claim to want.

Prior Tensions With The Government

This is not the first time Anthropic and the administration have found themselves in a public standoff. Earlier this year, Anthropic refused a Department of War request to remove safeguards covering domestic surveillance and fully autonomous weapons systems, a confrontation that resulted in the DoW threatening to label the company a supply chain risk. Sacks has specifically noted that the Fable situation is separate from that dispute, and that the administration still values Anthropic’s technical capabilities. But the pattern of a company that advocates for safety constraints and then resists government oversight when it arrives at an inconvenient moment is harder to ignore now that it’s happened twice.

A person close to the White House told Semafor that Amazon’s Andy Jassy had been in contact with administration officials about the jailbreak, suggesting the original warning came through the company’s cloud partnership with Anthropic. Amazon has not confirmed the details.

Where Things Stand

Sacks says the administration wants this resolved quickly and has framed it plainly: fix the jailbreak, lift the export control, get Fable back in market. “The ball is in Anthropic’s court,” he wrote.

Anthropic has described the situation as a misunderstanding it’s working to resolve. But the gap between the government’s account — that Amodei was asked directly and refused — and Anthropic’s framing of cooperative engagement and verbal-only evidence hasn’t closed yet. Until it does, both Fable 5 and Mythos 5 remain offline for essentially everyone.