Report: Amazon raised Anthropic jailbreak concerns before Fable cutoff

• The Information's reporting, surfaced by steph_palazzolo added a new named actor to the Fable shutdown: Amazon CEO Andy Jassy was among the tech leaders who raised Anthropic model concerns to Trump officials before the export-control order hit.steph_palazzolo's report
• Axios' follow-up, linked by theo, said Amazon called administration officials Thursday night with a jailbreak report on Mythos, then the White House moved toward Friday's takedown.theo's Axios confirmation theo's earlier report
• In Anthropic's official statement, the company said the directive barred access by any foreign national, including Anthropic employees, and that the only practical way to comply was to disable Fable 5 and Mythos 5 for everyone.rohanpaul_ai's summary
• Anthropic and the administration are describing the underlying issue very differently: Anthropic called the disclosed jailbreak narrow and non-universal in its statement, while kimmonismus citing David Sacks said the White House viewed it as a confirmed exposure of advanced cyber capability.kimmonismus on Sacks
• Buried in Anthropic's statement was a second reveal: Fable's safety posture already included 30-day customer data retention so the company could study and mitigate jailbreaks, a tradeoff that now sits at the center of the dispute.deredleritt3r's timeline

You can read Anthropic's statement on the directive, the first Axios scoop on the export restriction, and the later Axios report naming Amazon's role. The Information's report, flagged by steph_palazzolo, is the piece that put Andy Jassy in the chain. Anthropic's own writeup also sneaks in two unusually specific details: the company says it got the order at 5:21 p.m. ET, and says the government still has only given it verbal evidence of the jailbreak claim.

The first public version of this story was about a government order and a pulled model. theo's earlier report and the initial Axios scoop established that the administration had moved to block foreign access to Fable 5 and Mythos 5.

A day later, steph_palazzolo's report added the missing name: Amazon CEO Andy Jassy was among the tech leaders who raised concerns with senior Trump officials. Axios' follow-up, which theo's Axios confirmation amplified, separately said Amazon had reported the issue to the White House.

That matters because Amazon is not an outside critic here. As kimmonismus' reaction and MatthewBerman's reaction captured, the oddity is that one of Anthropic's biggest backers appears in the reporting as the party that helped trigger the crackdown.

Anthropic's own statement says the directive suspended access by any foreign national, whether inside or outside the US, including Anthropic employees. Because the company could not selectively enforce that in real time, it says the net effect was an abrupt global disablement of Fable 5 and Mythos 5.

That broad interpretation is why the shutdown looked so strange on first read. kimmonismus' thread described the rule as foreign-national access control that, in practice, disabled both models for everyone. giffmana's question also surfaced a still-open ambiguity: what exactly counts as Mythos or Fable for internal employee restrictions when the same capabilities may exist across adjacent tooling and internal systems?

The first Axios scoop framed this as a deemed-export style restriction on access to frontier models, not a country blocklist. That is a sharper precedent than a normal regional takedown, because the unit of control is the model itself plus the nationality of the person touching it.

Anthropic says the government's concern is a bypass of Fable 5. In the full statement, the company says the disclosed technique exposed only a small number of previously known, minor vulnerabilities, and says other publicly available models can find the same issues without any special bypass.

The administration side, at least in public retellings, is describing something harsher. kimmonismus on Sacks said David Sacks blamed Anthropic for refusing a request to fix a confirmed jailbreak, while ai_for_success's summary framed the White House demand as fix the jailbreak or pull the model.

That gap in language is the story. Anthropic says narrow, non-universal, and not Mythos-specific. The administration's allies say confirmed, dangerous, and serious enough to justify emergency action. levie's follow-up adds a useful calibration point from outside the White House framing: trying to jailbreak frontier models is normal practice, and sharing the results with the government is exactly what a security-testing partner would be expected to do.

By Friday night the dispute had become personal as well as technical. deredleritt3r's timeline stitched together the early reporting into a sequence: the government had approved Fable 5, another company then demonstrated a jailbreak, officials asked Anthropic to pause release, and Anthropic declined.

A later post from deredleritt3r's follow-up pushes that sequence further, claiming senior officials told Dario Amodei he was making a bad decision and that the conversation turned adversarial. That post is commentary, not primary reporting, but it lines up with the stronger sourced pieces in Axios' follow-up, which describes a White House scramble after Amazon's report.

The result is a rare picture of frontier-model governance happening by phone call first and formal process second. Anthropic says in its statement that it received the directive at 5:21 p.m. ET and that the letter itself did not provide specific details of the national security concern.

One of the more revealing details sits far down in Anthropic's statement. The company says Fable required 30-day retention of customer data so it could research and mitigate jailbreaks, and says that policy carried real customer costs even before the government intervened.

That detail connects two arguments Anthropic is making at once:

• perfect jailbreak resistance is not achievable today
• non-universal jailbreaks should be managed with defense in depth, monitoring, and fast mitigation
• the company wants government authority to block unsafe deployments, but only through a statutory process that is transparent and technically grounded

So the final twist here is that Anthropic was already treating Fable as a model that needed extra monitoring and data retention. The government looked at that same class of risk and concluded the model should be pulled from the market until the process around it changed.