The Fable 5 shutdown decoded: Amazon triggered it, Anthropic refused a patch, and now both sides negotiate

Fable 5 public availability

3 days

Ultimatum window given to Anthropic

90 min

Anthropic IPO target valuation

$96.5B

Enterprise partners affected

165K+

統計カードを読み込んでいます…

The Saturday-morning news cycle after the Fable 5 shutdown was less about the ban itself — that story broke Friday night — and more about the question nobody had answered yet: how did it actually happen? By Sunday afternoon, the answer was clear enough to be uncomfortable for several parties, including Anthropic's largest investor.

Amazon pulled the trigger

According to reporting by the Wall Street Journal, the chain of events that ended Fable 5's three-day public life began not at a government agency but inside Amazon.1

Amazon CEO Andy Jassy personally called Treasury Secretary Scott Bessent and other senior Trump administration officials. His company's researchers had obtained "information that could be used in cyberattacks" by using specific prompts with Fable 5. Jassy framed the call as a general security warning.2

The White House immediately convened an emergency meeting. What had begun as a partner flagging a concern quickly escalated into a Commerce Department ban.

The complexity here is notable. Amazon is Anthropic's primary cloud infrastructure provider, a major equity investor, and an enterprise customer deploying Claude across its own software-security business. It is also, now confirmed, the company whose research report triggered a federal export-control action against the AI it helped fund.2

Andrew Morris, founder of cybersecurity firm GreyNoise Intelligence, reviewed the technical specifics and said the information Amazon's researchers obtained is "quite far from truly dangerous cybersecurity intelligence" — comparable to output from many publicly available security tools.2 Anthropic itself had noted that the same vulnerabilities were discoverable from GPT-5.5 and other live models without requiring any bypass.3

A separate thread exists: the UK's AI Safety Institute reported June 9 — the same day Fable launched — that its cybersecurity team had developed a jailbreak "within a few hours of access." AISI's Xander Davies wrote that the technique was extended over two additional days to allow "multiple steps of malicious agentic tool-calls."4 It is not confirmed whether the AISI findings or Amazon's report drove the specific directive — both may have contributed.

A 90-minute ultimatum and a refusal

A source close to Anthropic told The Hill that on Friday the administration told the company it had 90 minutes to pull Fable 5, without providing prior notice of any national security concern.5 Amodei and senior Anthropic personnel were on the phone with the White House within 15 minutes. The administration provided no further details.

David Sacks, co-chair of the President's Council of Advisors on Science and Technology, gave a different account on Saturday. He said the administration had asked Amodei last week to either patch the vulnerability or de-deploy Fable 5 — and that Amodei refused.6 According to Sacks, Anthropic's defense — that the jailbreak isn't serious because it exists in competing models — "is very much at odds with their branding and ethos as a safe AI research firm."5

Sacks added that Trump signed the export-control order "reluctantly," that the action was intended to be temporary, and that the ball is "in Anthropic's court."6

コンテンツカードを読み込んでいます…

Anthropic maintained its position through the weekend: the vulnerability cited is minor, demonstrable in several other frontier models without requiring any bypass, and does not meet the threshold justifying a unilateral model recall. The company says it worked with government AI testing agencies on a phased pre-release process and received explicit approval to deploy Fable 5.5

Hegseth, Sacks, and a divided Washington

The political fallout has not followed a clean partisan line, which complicates Anthropic's calculus heading into negotiations.

Defense Secretary Pete Hegseth posted on X Saturday: "Three months ago, @DeptofWar kicked @AnthropicAI out of our building — forever. Every passing day proves why that was the right move."5 Pentagon CIO Kirsten Davies added: "Some things are simply more important than revenue cycles, clickbait, and pre-IPO valuation. America First."6

The administration's hostility predates this incident. Hegseth designated Anthropic a supply-chain risk earlier this year after the company refused to allow the Pentagon to use Claude for fully autonomous lethal weapons or mass domestic surveillance. Anthropic has been suing the government over that designation; a San Francisco federal judge granted a preliminary injunction pausing the directive.7

On the other side of the aisle: Democratic Senator Mark Kelly of Arizona told CBS News's Face the Nation on Sunday that the export-control directive was "the right call" and that AI companies should be "incredibly careful" with model releases. He said Anthropic "seems to be willing to work with the federal government."8 The bipartisan support for the order is a signal Anthropic will need to weigh carefully in any public push-back strategy before its IPO.

The pre-IPO context is acute. Anthropic filed an S-1 with the SEC on June 1, targeting a public listing in late summer or fall with a reported $96.5 billion valuation.9 Losing its two most capable models — and accruing regulatory conflict — is the worst possible preparation for a roadshow.

Allies push back — UK, EU, and a geopolitical rupture

The international reaction to the global shutdown (not just the foreign-nationals restriction, but the full all-users takedown) has been notably pointed.

UK Prime Minister Keir Starmer is personally lobbying the Trump administration to allow British users a carve-out — to be placed on an approved-country list that would restore Fable 5 and Mythos 5 access. Downing Street told The Telegraph it is "hopeful" the shutdown will prove temporary.10 British MP Tom Tugendhat noted that dozens of UK hospitals and research institutions had their access cut off overnight, characterizing the incident as evidence that "sovereignty is more about code than cannons."4

The European Commission said it is "assessing the practical consequences" and issued a notable warning: contingency measures "should not be discriminatory against partners."4 French politicians Jordan Bardella and Bruno Retailleau went further, calling the ban a "wake-up call" and urging European governments to accelerate backing of homegrown alternatives — with Mistral AI cited explicitly.4

This is the first time a U.S. export-control action has had allied governments formally lobbying for individual carve-outs on a commercial AI product. The precedent — and the friction it introduces into U.S.-EU and U.S.-UK tech relations — may outlast the specific dispute over Fable 5.

Customers demand refunds as enterprise deployments stall

For paying subscribers, the shutdown produced a concrete financial question: do they get money back?

Fable 5 was priced at roughly twice the per-token rate of Claude Opus 4.8 and was included in Claude subscription plans through June 22. Many users upgraded accounts specifically to access it after the June 9 launch. With access cut on June 12, Anthropic sent emails offering a "prorated refund" to users who purchased or upgraded between June 9 and June 14 — with a deadline to cancel by June 20.7

Reports on Reddit suggest the refund process is inconsistent: some customers say they cancelled and received nothing. EU-based customers appear to have had more success invoking the EU Consumer Rights Directive's 14-day cooling-off period via Anthropic's in-app support.7

At the enterprise level, the impact hits Anthropic's newest partners hardest. TCS (50,000 employees across 56 countries) and DXC Technology (115,000 employees, 70 countries) both signed Global Premier partnerships in the days before the ban and had plans to deploy Fable-class capabilities to their clients in banking, government, and critical infrastructure. Those deployments are now on hold.

Anthropic negotiates for restoration

By Saturday night, Anthropic was in active discussions with the Commerce Department to restore access.9 Sacks said the administration wants the issue resolved quickly and framed the path forward as straightforward: Anthropic patches the vulnerability, the export control is lifted, Fable returns to general release.

Anthropic's counter-position is more complicated. The company believes the jailbreak cited does not meet the technical threshold for a model recall and has said so publicly. Accepting the government's framing — that a patch is necessary — would validate a regulatory precedent it has actively opposed: that a government can require a unilateral model withdrawal based on a non-universal, non-novel security finding that exists in competing products.

That tension will likely define the negotiation. Anthropic needs Fable 5 back online — for revenue, for enterprise relationships, and for its IPO narrative. But how it comes back, and what it concedes in the process, will shape how the next disagreement with Washington unfolds.

Watch for: a joint technical statement, a time-limited remediation agreement, or a court motion expanding the scope of Anthropic's existing Pentagon lawsuit to cover the Commerce Department directive.

A government-style control room with multiple monitors and a control panel, symbolizing regulatory oversight of AI systems — Control room — Pexels via Pixabay (royalty-free)