← Back to feed

Six days into the Fable 5 shutdown, the political framing of the ban has shifted materially. David Sacks, co-chair of the President’s Council of Advisers on Science and Technology, disclosed on June 13 that the Trump administration had offered Anthropic a choice before issuing the export control directive: fix the jailbreak vulnerability, or voluntarily de-deploy the model. According to Sacks, Dario Amodei refused both options. The government then forced the shutdown itself.

That sequence reframes everything. Anthropic’s public statement after the shutdown described the government’s action as a misunderstanding and noted that the same vulnerability exists across other frontier models. That argument lands differently once the record shows Anthropic was offered the chance to patch first and declined.

What Happened, In Order

The jailbreak in question affects Fable 5 and Mythos 5. The specifics have not been made public, but UK AISI confirmed in Glasswing reporting that Mythos can solve advanced cyber ranges end-to-end — capability that informed the government’s concern. Anthropic’s position, as reflected in its public statement, is that the jailbreak is an industry-wide problem, not a Fable-specific one, and that forcing a shutdown selectively punishes a US company without addressing the underlying issue.

The White House’s position, now stated on the record by Sacks, is that Anthropic was given a meaningful off-ramp and chose neither path available to it. That makes the shutdown less an arbitrary policy intervention and more the consequence of a specific negotiating stance.

Anthropic dispatched senior engineers to Washington on June 16 for in-person talks with Commerce Department officials. It is the first face-to-face meeting since the directive was issued on June 12. Observers describe the session as deal-seeking rather than compliance-focused, suggesting both sides want a path to restoration.

What Markets Are Pricing

As of June 17, prediction market participants are pricing restoration as the more likely outcome:

• Polymarket: 67% odds that Fable 5 access is restored for US customers before July 1
• Kalshi: 58% odds by July 1, rising to 74% by July 10
• More than $544,000 has traded on Polymarket’s contract

The spread between the two platforms (67% vs 58% by the same date) reflects uncertainty about the pace of any deal, not about the direction. Both markets expect restoration.

What a Deal Might Look Like

The structure of any agreement is not public, but the Sacks disclosure narrows the options. The original two choices — patch or withdraw — remain on the table. A third path would involve some combination: Anthropic commits to a remediation timeline, the government issues a time-limited exemption, and Fable 5 returns under monitoring conditions similar to those governing Mythos access for Glasswing partners.

Whether Anthropic’s engineers left Washington with anything resembling a framework is not known. No outcome from the June 16 meeting has been disclosed publicly.

The Broader Precedent

The Fable 5 case is the first time a US government has forced a frontier AI model offline on national security grounds. The mechanics of how this plays out — whether a patch is sufficient, whether the government accepts vendor attestation or demands independent verification — will set precedent for every lab operating at the frontier. That includes OpenAI, Google DeepMind, and xAI, all of which have models with documented jailbreak surfaces.

Anthropic has, by refusing to patch on government request, made itself the test case. If it wins on terms, other labs benefit from the precedent. If it loses, the regulatory environment for frontier model deployment in the US just changed permanently.