A White House adviser says Anthropic refused to fix a cyberweapon jailbreak and got its model banned for it. Anthropic says the flaw is minor and reproducible anywhere. Almost every fact that would settle it is non-public — and that is the real story.
Over the weekend, White House AI adviser David Sacks — co-chair of the President’s Council of Advisors on Science and Technology — published the most detailed government account yet of why Washington pulled the plug on Anthropic’s most powerful models. It is a striking document, and it flatly contradicts Anthropic’s own version of events.
I have been openly critical of Anthropic in this space — of the way its safety narrative doubles as a competitive moat. Consistency requires that I bring the same skepticism to a government official’s account justifying an unprecedented intervention on the strength of evidence none of us can see. This piece is not about whose side to take. It is about the fact that “safety” has become a card every powerful party in this fight is now playing — and that the public has been handed no way to check any of them.
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.
What Sacks claims
Sacks frames his post as what he believes to be true after conversations inside and outside government. Stripped to its load-bearing assertions, his account runs like this.
Fable, he says, is simply Mythos with guardrails — and if those guardrails fail, you have handed Mythos-class cyber capability to people who should not have it. He pointedly reminds readers that Anthropic itself promoted Mythos as something close to a cyberweapon and asked for it to be regulated as one, so a failure of Fable’s safeguards is, by Anthropic’s own logic, Anthropic’s responsibility to fix. A “highly credible” partner trusted by both Anthropic and the government, he says, was testing Fable and surfaced a jailbreak of those guardrails. The administration asked Dario Amodei to patch it or pull the model; according to Sacks, Amodei refused. Only then, and reluctantly, did the administration issue the export control.
He goes further, and this is the sharpest part: he rejects Anthropic’s characterization of the jailbreak as not serious, calling that minimizing language inconsistent with the company’s brand as the safety lab, and arguing it is hard to see how a bypass that restores the operability of a cyberweapon could be called anything but serious. Anthropic, in his telling, chose to keep its consumer model live rather than prioritize safety. He says the administration wants the control lifted as soon as the issue is remediated, denies the move has anything to do with Anthropic’s earlier disputes with the Pentagon, and closes with a line built for the timeline: the ball is in Anthropic’s court.
Privacy Tools in the Age of AI: Practical Strategies with VPNs, Secure DNS, Private Relay and Intelligent Defenses (Build Your Own VPN)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Anthropic says
Anthropic’s account, given in its June 12 statement and repeated since, describes a very different object. The company says US authorities provided no specific technical detail, and that its own understanding is that the government believes it found a way to bypass a safeguard meant to stop the model from being used to find software vulnerabilities. It reviewed a demonstration of the technique, it says, and what it saw was the identification of a small number of previously known, minor flaws — results that other public models, including OpenAI’s GPT-5.5, can produce without any bypass at all. On that basis Anthropic disagrees that a “narrow potential jailbreak” should force the recall of a model used by hundreds of millions, and warns that applying such a standard across the industry would halt frontier deployments entirely. It apologized to customers, said it disabled both models worldwide only to comply with the order, and restated that it supports government authority to block unsafe deployments solely through a process that is transparent, fair, and technically grounded.
Cybersecurity for Business: Organization-Wide Strategies to Ensure Cyber Risk Is Not Just an IT Issue
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The two stories cannot both be the whole truth
Set the accounts side by side and the contradiction is not about tone. It is about the thing itself. One side describes the restored operability of a cyberweapon; the other describes a parlor trick that finds already-known bugs and works on competitors’ models too. Those are not two framings of one fact. They are two different claims about how dangerous the bypass actually is — and at least one of them is substantially wrong.
Here is the problem for everyone reading along at home: you cannot tell which. The jailbreak is not described. The vulnerabilities are not named. There is no CVE, no published methodology, no independent assessment. The “highly credible trusted partner” is unnamed. We are asked to choose between a government official’s secondhand summary and a vendor’s self-interested minimization, with no instrument to weigh either. On a question of national-security consequence, the entire evidentiary record is a matter of trust in parties who each have a reason to shade it.
Detekt® Indoor Air Quality Test Kit – 6 Mold + 6 Bacteria Test – Home/HVAC
- Made in USA: Trusted quality and customer service
- Includes Species Guide & Consultation: Over 3x species coverage and expert help
- Multi-Location Testing: Tests indoor air, surfaces, and HVAC
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The detail both sides are quieter about
There is one more actor, and it complicates Sacks’s “trusted partner” framing considerably. According to reporting by Semafor, picked up by Fortune and others, the entity that flagged the jailbreak to the government was Amazon — and Amazon CEO Andy Jassy was reportedly in contact with the administration about it. Amazon has not confirmed the specifics, saying only that governments often seek its counsel on security risks.
Sit with that. Amazon has poured billions into Anthropic and supplies much of the cloud Anthropic runs on. It is simultaneously a competitor whose own models vie with Claude, and an investor with a direct stake in how, when, and on what terms Anthropic’s most capable systems reach the market. A “highly credible trusted partner” is one description of that relationship. “A rival and stakeholder whose interests are not neutral” is another, equally accurate one. This does not mean the jailbreak is fake or that Amazon acted in bad faith — flagging a real cyber risk is exactly what a responsible partner should do. It means the clean picture Sacks paints, of a disinterested tester surfacing a danger and a government acting only on the merits, has a commercial undercurrent that his post leaves out.
7-in-1 Hidden Camera Detectors, AI Chip Anti-Spy Camera Finder, 6 Modes
- AI-Enhanced Detection: Upgraded AI chip for fast, accurate scans
- Adjustable Sensitivity: 5-level sensitivity for precise device locating
- Long Detection Range: Detects signals up to 27 feet away
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Everyone is holding the same card
Step back and the pattern is almost too neat. The government invokes safety to justify the most forceful intervention it has yet made in commercial AI. Anthropic, which built the entire “this is a cyberweapon, regulate it” framing, now finds that framing aimed at its own product — and suddenly argues the danger is overstated. And a rival-investor may have invoked safety in a way that conveniently kneecaps a competitor’s flagship launch. Each party’s safety claim points in the direction of its own advantage.
I wrote recently that the safety state, once built, would not belong to Anthropic. This is that thesis arriving faster than I expected — with a twist. The off-switch Anthropic spent years arguing for got built, and the first time it was thrown, it was thrown at Anthropic, possibly on a tip from the company’s own backer. But the lesson is not that the government is the villain, or that Amazon is, or even that Anthropic is. It is that “safety” has become the universal justification in AI — a word that licenses a ban, a denial, or a competitive strike depending on who is holding it — and the more load-bearing the word becomes, the less anyone is required to show their work.
The accountability gap
The honest conclusion is uncomfortable because it does not let you root for anyone. A government can now suspend a product used by hundreds of millions, worldwide, on the basis of a classified finding, an unnamed partner, and an undisclosed technique. A company can wave away a national-security concern as trivial without showing the evidence that would prove it. And a third party with tangled incentives can put a thumb on the scale, off the record. None of these actors is offering the one thing that would let the rest of us judge: a transparent, technically grounded, independently reviewable process.
Which is, notably, exactly what Anthropic says it wants — and exactly what would also constrain Anthropic. That is the standard worth demanding, and the reason to demand it is not loyalty to any party in this fight. It is that the alternative is the world we are in now, where the most consequential decisions in AI are made on secret evidence and adjudicated in dueling press statements.
The honest read
If you came here expecting me to side with the safety lab against the administration, or with the administration against the lab, I cannot help you, because the evidence to do so honestly does not exist in public. What I can say is this. The same skepticism I apply to Anthropic’s safety branding, I apply to a political off-switch justified by evidence I am not allowed to see and surfaced, in part, by a competitor. The Fable standoff is not a morality play with a hero. It is a preview of how power over frontier AI will actually be exercised: through control of a chokepoint, in the name of safety, by whoever has their hand on it that week.
Watch what gets resolved, and how. If Anthropic quietly ships a patch and the ban lifts within days, the “minor, reproducible anywhere” story looks thin. If the standoff drags, the “trivial flaw” defense looks more credible and the intervention looks more like leverage. Either way, demand the methodology, not the press release. The party that benefits from your trust is exactly the party that should have to earn it.
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 public 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. © 2026 Thorsten Meyer · Powered by Thorsten Meyer AI. See Imprint/Impressum and Privacy Policy.
Sources and further reading
- David Sacks — statement on X (June 13, 2026)
- Anthropic — Statement on Fable 5 and Mythos 5 access
- Tom’s Hardware — Trump adviser David Sacks says Anthropic refused to fix Fable 5 jailbreak
- Fortune — How a warning from Amazon led the White House to shut down Anthropic’s Mythos model
- Benzinga — Sacks: US wants to lift Anthropic export controls “as soon as possible”
