TL;DR
White House AI adviser David Sacks said Washington restricted Anthropic’s Fable S models after a trusted partner found a jailbreak that could restore Mythos-level cyber capabilities. Anthropic disputes that account, saying the government shared no technical detail and that the issue was minor and reproducible in other public models.
White House AI adviser David Sacks has said the government restricted Anthropic’s Fable S models after a trusted partner found a jailbreak that could restore cyberweapon-like capabilities, while Anthropic says the cited flaw was minor and not enough to justify recalling a widely used AI model.
In a June 13 statement on X, Sacks described Fable as Mythos with added guardrails and said the administration asked Anthropic Chief Executive Dario Amodei to fix the bypass or pull the model. Sacks said Amodei refused, leading officials to issue an export control that he described as reluctant.
Anthropic, in a June 12 blog post, gave a different account. The company said the government supplied no specific technical detail, that the demo showed only a few minor flaws already known to researchers, and that similar behavior could be reproduced in other public models, including GPT-5.5, without a special bypass.
The competing accounts leave the core technical question unresolved: whether the bypass reopens Mythos-class cyber capability, as Sacks claims, or whether it is a narrow failure common across current systems, as Anthropic claims. Semafor, in reporting carried by Fortune and other outlets, identified Amazon as the possible partner that flagged the issue, but Amazon has not confirmed that role.
The Safety Card, Played From Every Side
● ContestedA White House adviser says Anthropic refused to fix a cyberweapon jailbreak and got banned for it. Anthropic says the flaw is trivial. Almost every fact that would settle it is non-public — and “safety” is now the card every side is playing.
Both are claims, not findings. They don’t disagree on tone — they disagree on what the bypass actually is.
- A “highly credible trusted partner” found a jailbreak of Fable’s guardrails.
- The admin asked Amodei to fix it or pull the model. He refused.
- So the export control was issued — “reluctantly.”
- It restores operability of a cyberweapon; calling that “not serious” is indefensible.
- The government gave no specific technical detail.
- The demo found a few minor, already-known flaws.
- Other public models (incl. GPT-5.5) do the same without a bypass.
- A “narrow potential jailbreak” shouldn’t recall a model used by hundreds of millions.
Per reporting by Semafor (carried by Fortune and others), the entity that flagged the jailbreak was Amazon — with CEO Andy Jassy reportedly in contact with the administration. Amazon hasn’t confirmed specifics. Flagging a real risk is what a good partner does — but Amazon wears three hats at once, and none of them is neutral.
Each actor’s safety claim points toward its own advantage.
The entire evidentiary record is a matter of trusting parties who each have a reason to shade it.
A transparent, technically grounded, independently reviewable process — which is, notably, exactly what Anthropic says it wants, and exactly what would also constrain Anthropic. The reason to demand it isn’t loyalty to anyone; it’s that the alternative is decisions made on secret evidence and adjudicated in dueling press statements.
Independent commentary, produced with AI assistance under human editorial oversight; the views are the author’s own and may change. This is analysis and opinion, not investment, financial, legal, or technical advice, and it concerns an actively developing situation in which key facts are disputed and non-public. Claims attributed to David Sacks reflect his June 13, 2026 statement on X; claims attributed to Anthropic reflect its published statements; reporting on Amazon’s role reflects accounts published by Semafor and others — all read as of June 15, 2026, and presented as the claims of those parties, not as established fact. Characterizations are the author’s interpretation, offered in good faith and open to rebuttal. References to specific people, companies, and government actions are factual and analytical, not partisan, and imply no affiliation or endorsement.
Secret Tests Drive AI Controls
The dispute matters because it places a major AI access decision on evidence the public cannot review. If the government’s account is correct, the action shows Washington is willing to block a commercial frontier model when officials believe safety controls have failed in a high-risk domain.
If Anthropic’s account is closer to the facts, the same action could show how opaque safety claims can be used to halt a product based on evidence that outside researchers, customers and rivals cannot examine. Either outcome affects developers, cloud providers, enterprise buyers and governments that rely on model access for security work and other high-value tasks.
AI Safety and Security: Architectural Context, Perspectives, and Insights
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Mythos To Fable S
Anthropic had previously framed Mythos as a model with dangerous cyber capability and called for regulation of systems in that class, according to the accounts summarized in the ThorstenMeyerAI Dispatch. Sacks used that history to argue that Fable S should be treated as a guarded version of Mythos, not as a separate low-risk release.
That history is part of why the dispute has wider force. Anthropic has built much of its public case around AI safety, while the government is now using a safety rationale against Anthropic’s own flagship release. Amazon’s reported role adds another layer because it is an Anthropic investor and cloud provider, while also backing models that compete in the same market.
“It restores operability of a cyberweapon; calling that not serious is indefensible.”
— David Sacks, White House AI adviser, on X
Key Evidence Stays Nonpublic
Several facts that would settle the dispute remain unavailable: the exact prompt or method used in the jailbreak, the test conditions, the identity of the trusted partner, the government’s risk standard, and any independent review. There is also no public CVE, no published methodology and no shared technical report that would let outside experts compare the two accounts.
Amazon’s alleged role is also unresolved. Semafor’s reporting says Amazon may have flagged the jailbreak, but Amazon has not publicly confirmed the specific claim in the source material, and the government has not named the partner.
Patch Timing Becomes Evidence
The next marker is whether the restriction is lifted after a patch, review or new technical filing. A quick end to the ban after a quiet fix would weaken Anthropic’s claim that the flaw was minor; a longer standoff would put more pressure on the government to release a reviewable basis for the action.
For now, readers should treat both accounts as claims. The case is developing as of June 15, 2026, and the strongest public test will be whether officials, Anthropic or an independent reviewer release enough detail to show what failed, how serious it was and why the export control was used.
Key Questions
What did David Sacks say happened?
Sacks said a trusted partner found a jailbreak in Fable S guardrails, the administration asked Anthropic to fix or pull the model, and Amodei refused. That remains Sacks’s account, not a public technical finding.
How did Anthropic respond?
Anthropic said the government provided no specific technical detail, that the flaws were minor and already known, and that similar behavior could be reproduced in other public models without a special bypass.
Is Amazon confirmed as the partner?
No. Semafor reporting carried by Fortune and other outlets identified Amazon as the possible partner, but Amazon has not confirmed that specific role and the government has not named the entity.
Why is this bigger than one model?
Safety claims are now being used by governments, model labs and commercial partners in decisions that can remove model access. Without public methods, readers and customers cannot tell whether a ban is risk control, market pressure or both.
What would help settle the dispute?
A technical report with the test method, affected model versions, severity criteria, remediation record and independent review would give outside experts a basis to compare the accounts.
Source: Thorsten Meyer AI
