David Sacks makes Washington's case against Anthropic

The standoff over Anthropic’s most capable models now has a public face on the government side. David Sacks — the Trump administration’s former AI and crypto czar, now a co-chair of the President’s Council of Advisors on Science and Technology — has laid out, in detail, the administration’s account of why it forced Anthropic to pull Fable 5 and Mythos 5. It is a markedly different story from the one Anthropic told in its own blog post.

I’ve had a number of conversations with folks inside and outside government about the current situation with Anthropic, and here is what I believe to be true… The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused… Anthropic prioritized the continued offering of the consumer model over safety… The ball is in Anthropic’s court.

— David Sacks (@DavidSacks) June 13, 2026

The administration’s account

In his telling — which tracks the version an administration official gave Axios — the sequence ran like this:

• Fable is Mythos with guardrails. Mythos is the powerful model; Fable is the public version with safety limits bolted on. If the guardrails fail, Sacks argues, you have exposed Mythos’s “advanced cyber capabilities” to people who should not have them — and Anthropic itself spent months casting Mythos as something close to a cyberweapon that needed regulation.
• A trusted partner found a jailbreak. “A highly credible trusted partner of both Anthropic and the USG who was testing Fable came forward with a jailbreak of those guardrails,” Sacks wrote. The administration, he says, asked Dario Amodei to fix the jailbreak or de-deploy the model — and “Dario refused.”
• The two sides disagree on severity. Anthropic’s blog called the flaw minor; Sacks says that is “not what the trusted partner and the USG believe,” and not “consistent with Anthropic’s brand as the AI safety company.”
• The control was a last resort. “In reaction, the Admin issued the export control. The Admin did this reluctantly,” he wrote, adding that it was “very surprised that Anthropic hasn’t wanted to cooperate with a reasonable safety request.”

Sacks also moved to head off a particular reading of events: those tying the action to the earlier Anthropic–Department of War dispute, he said, “are wrong.” The administration, in his account, “values Anthropic’s technical capabilities” and considers the issue “serious” but “easily resolved.”

Anthropic’s side

Anthropic’s published position is the mirror image. It describes what the government demonstrated as a “narrow, non-universal jailbreak” that would unlock Mythos’s cyber capabilities in one specific instance, not a universal break of Fable’s defences, and says the capability is “widely available from other models.” Its core objection: “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.”

So this is not merely a fight about what to do next. The two sides cannot agree on how bad the underlying flaw is — and that gap, more than the export control itself, is what will determine how long this lasts.

The context that colours it

Worth weighing alongside the account is the messenger. Sacks has been one of Anthropic’s sharpest critics inside the administration, having accused the company of “regulatory capture” and “fear-mongering” over its push for AI rules. None of that disproves his version of events, but it is the backdrop to a dispute now being argued out in blog posts and X threads rather than behind closed doors.

What it signals

The striking thing is that both sides claim to want the same ending: Fable back in general release once the flaw is fixed. Sacks says the administration’s “hope now is that Anthropic remediates the safety issue, the export control is lifted, and Fable goes back into general release,” and that it wants this “as soon as possible.” That framing suggests a resolvable spat rather than a terminal break.

What to watch over the coming days:

• Whether Anthropic patches and the control lifts. Sacks says that is the path; a fix plus a lifted order would prove it.
• Whether “Dario refused” is contested. Anthropic has not used that framing; if it disputes the sequence, this gets longer and uglier.
• The precedent. A lab that built its brand on safety and the White House’s AI point man now publicly disagree over whether a flaw is “serious” — while a US frontier model sits switched off worldwide. For anyone outside the US, the lesson from the original ban holds: access to a frontier model can hinge on an argument you have no seat in.