The weekend's Anthropic-vs-White House standoff over Fable 5 grinds into Monday with no resolution — and NLW walks through the dueling press releases, the Amazon jailbreak report, the wellness-retreat meme, and the dawning realization that this fight was never going to be settled by engineers.
The Fable 5 fight won't be resolved by engineers — it's interpersonal.
A jailbreak report from Amazon escalated into a Commerce Department export ban that forced Anthropic to pull Fable and Mythos entirely. But the more NLW digs into the dueling accounts, the clearer it gets: the technical merits barely matter. The administration felt Anthropic wasn't taking it seriously; Anthropic thought it could simply reason the White House into agreement. It can't. Anthropic is no longer a scrappy startup — it's one of two leaders in the most consequential industry of the era, and it has to play ball with the government it has, not the one it wishes it had.
— David Sacks, in a public post. Former AI czar David Sacks published a lengthy account framing Anthropic as the holdout: a trusted partner found a jailbreak of Fable's guardrails, the admin asked Dario to fix it or pull the model, and he refused. Sacks cast the export control as a reluctant last resort.
NLW notes Sacks paints Anthropic as hypocritical — the safety company not taking a safety issue seriously — and pointedly shifts from 'Anthropic' to naming Dario Amodei personally. No resignation calls yet, but singling out Dario at least hints at a second way out of the standoff.
— Eric Voorhees. AI entrepreneur Eric Voorhees offered the skeptics' counter-narrative: anyone who's received bug reports knows the type, Anthropic thought the halt order was absurd, and the feds seized on prior grievances. 'Anthropic has more credibility on such topics than Washington.'
Anthropic argued the bypass shared with them was specific and discrete, not a universal jailbreak that strips all guardrails — and that perfect jailbreak resistance isn't possible today. NLW notes the guardrails were so broad they blocked questions about mitochondria and any prompt with the word 'cancer,' so a 'jailbreak' could be trivial or catastrophic.
Multiple outlets identified Amazon as the company that reported the jailbreak to the government Thursday night, showing how it accessed portions of the Mythos model. Anthropic says it notified the government multiple times before the June 9 release with no objections.
The Wall Street Journal reports the decision to shut Fable down followed conversations between Amazon CEO Andy Jassy and Treasury Secretary Scott Bessent. Officials fielded calls from at least five other companies, but reporting suggests the ban rested almost solely on Amazon's report.
— Andrew Morris, founder of GrayNoise Intelligence. GrayNoise Intelligence founder Andrew Morris said Amazon's researchers got Fable to discuss bugs in at least four platforms — normally blocked — but that the info was far from dangerous and available from many other models. The unique risk would be turning vulnerabilities into working exploit code, which researchers reportedly never demonstrated.
Anthropic says it got a 1pm call, was told it had 90 minutes to take Fable and Mythos down over a 'national security threat,' but received no specifics. Formal export-control notice came at 5:30pm; the models went dark around 10pm.
— The Wall Street Journal. Trump's name was conspicuously absent from the list of decision-makers, which included Bessent, Chief of Staff Susie Wiles, Cyber Director Sean Cairncross, and Commerce Secretary Howard Lutnick. NLW argues Trump now looks like the only primarily innovation-concerned actor left in the White House.
A senior White House official told Politico export controls were 'a last resort after begging them for hours to work with us.' A source close to Anthropic flatly disputed it: 'There was never any begging or asking us to work with us, just a declared ninety-minute deadline.'
— A White House official, to Politico. A Politico White House source argued that had Anthropic moved to fix or pause access rather than dismissing the report as isolated, the ban never would have happened. Anthropic's calm, rational explanation read in the room as not taking the threat seriously.
— Cory Ward. Software engineer Cory Ward argued the guardrails were never meant to stop finding or fixing bugs, only to prevent identifying and weaponizing new exploits. 'There is nothing for Anthropic to actually resolve... It is entirely down to politics.'
— Miles Brundage. AI policy researcher Miles Brundage suggested officials thought something unsurprising was surprising — and that no domain experts at CAISI or the NSA appear to have been looped in. Colin Kremerer added that many knowledgeable White House tech people have left, leaving the room out of its depth.
White House sources said Amodei was unreachable at a wellness retreat; Anthropic and reporter Ashley Vance, who was at HQ that day, called it false. Vance: 'The feds seem to be scrambling to try and make an example of Anthropic again. This is not technical, it's petty.'
— Jeff Cafe. Jeff Cafe argued the 'wellness retreat' line was a sticky, image-generating idea no one can unsee — Dario in a bathrobe with cucumbers on his eyes as someone delivers the news. A reminder that the fight is being waged in memes as much as memos.
Semafor reported the export controls were imposed partly over suspicions a China-linked group had accessed Mythos, citing 'a person familiar.' The article carried almost no detail, and Anthropic said the White House never raised China in discussions and that its models are already blocked there.
— Pete Hegseth, Defense Secretary. Defense Secretary Pete Hegseth's tweet added fuel to the theory that the crisis is personal — despite Sacks insisting it isn't. NLW's point: even if making consequential policy on the basis of who likes whom is abhorrent, that's the world Anthropic now has to operate in.
— Ben Thompson, Stratechery. Ben Thompson's 'Anthropic's Safety Superpower' argued every contested Fable decision traces to safety because Anthropic genuinely believes it alone takes superintelligence seriously. He respects how effective that conviction is — and fears people convinced they know what humanity needs building technology that could rival nation-states.
NLW says he's seen this dynamic up close: the moment people convince themselves they're the only ones sufficiently concerned and the only ones with the right fix, danger follows — and the stakes are radically amplified by the power inherent in superintelligence.
— Adam Thierer, RSI senior fellow. RSI's Adam Thierer argued that whatever got us here, the policy is disastrous on the merits: a leading US AI company forced to pull a product millions used over non-public, unexplained concerns. He framed it as a major escalation in centralizing control over advanced computation — by an administration that called winning the AI race a priority.
The WSJ reports Anthropic sent senior technical staff — Nicholas Carlini, risk-evaluation lead Logan Graham, and head of safeguards David Orr — to meet government security experts and de-escalate. Separately, cybersecurity leaders led by ex-Facebook CSO Alex Stamos published an open letter urging Lutnick and Cairncross to lift the directives.
NLW's read: the resolution won't be primarily technical, it'll be interpersonal. Anthropic thought it could reason the White House into its view; it can't. Investor Melinda Chu put the bottom line bluntly — sending senior staff won't matter unless Amodei himself shows up.