Amazon Just Killed Anthropic’s Best Model. From the Inside.
In a nutshell
The story of Anthropic's June 12 shutdown was not, as initially reported, primarily about a jailbreak. It was about a phone call.
Three reputable news organisations have now confirmed the same set of facts. The Wall Street Journal reported it first. The Information confirmed it independently. Reuters added details. The Claude Fable 5 worldwide shutdown was triggered by Amazon — Anthropic's largest investor, board member and primary cloud host.
This is the story that changes how every European enterprise must think about AI vendor risk.
The Phone Call
According to the Journal's reporting, confirmed by The Information and Reuters, Andy Jassy — Amazon's CEO — personally contacted Treasury Secretary Scott Bessent and other senior officials on Thursday, June 12, 2026, to relay findings from Amazon's own security researchers: that Anthropic's newly launched Claude Fable 5 model could be prompted, through a specific technique, to produce information useful for conducting cyberattacks.
Those conversations directly preceded the export control directive issued by Commerce Secretary Howard Lutnick to Anthropic CEO Dario Amodei at 5:21 PM ET that Friday, ordering the company to suspend access to both Fable 5 and its more powerful restricted sibling, Mythos 5, for all foreign nationals worldwide — including Anthropic's own non-US employees. Anthropic, unable to verify user citizenship at the API level in real time, disabled both models for every customer globally.
Read that sequence again. Amazon's CEO called the US Treasury Secretary on Thursday. The US Commerce Secretary ordered Anthropic's models offline on Friday at 5:21 PM ET. Twenty-four hours. From private corporate phone call to global regulatory action.
The Conflicts of Interest — All of Them
Amazon's relationship with Anthropic is layered with conflicts of interest that any reasonable regulatory framework would recognise as disqualifying for a security-tip-line role.
Conflict one — Investor. Amazon has committed approximately $8 billion to Anthropic across two investment rounds — making it Anthropic's largest single investor. Amazon's $50 billion OpenAI partnership announced in February 2026 created the first competitive tension. Amazon is now an investor in two of the world's most consequential AI companies — one of which it just helped regulate into a multi-day product outage.
Conflict two — Board member. Amazon has board representation at Anthropic. The board's role is to act in the best interest of the company. Triggering a regulatory action that takes the company's most capable products offline globally is not normally considered acting in the company's best interest.
Conflict three — Cloud host. Amazon Web Services hosts Anthropic's models. AWS revenue from Anthropic's compute consumption is significant — and growing. An AWS spokesperson noted that AWS itself was affected by the model cutoff. Amazon caused damage to its own cloud business by triggering this action.
Conflict four — Competitor. Amazon's $50 billion OpenAI partnership makes OpenAI Amazon's primary AI distribution partner. Anthropic competes with OpenAI. A regulatory action that takes Anthropic's frontier models offline benefits OpenAI directly — and benefits Amazon's OpenAI partnership directly.
Four conflicts. One phone call. Twenty-four hours to global product shutdown.
What This Means for Every AI Vendor Relationship
Every European enterprise that has selected an AI vendor based on capability, safety credentials or commercial terms now faces a question that did not exist on Thursday: what happens if your AI vendor's largest investor calls a US cabinet secretary about your vendor's product?
The Anthropic case answers that question definitively. The answer is: the product can be taken offline globally within 24 hours, with no warning, with no due process and no European recourse.
This is not a theoretical risk. It is now precedent.
The Anthropic Response — Constrained Compliance
Anthropic published a statement within hours of the directive. The company called the action a likely misunderstanding and said it is working to restore access quickly. Anthropic's emphasis on compliance while gently pushing back indicates a desire to maintain constructive dialogue.
Read the language carefully. "Likely misunderstanding." "Working to restore access." "Constructive dialogue." This is the language of a company that cannot afford to publicly attack its largest investor — even when its largest investor has just triggered a regulatory action that disabled its flagship products worldwide.
Anthropic's actual position — that the jailbreak cited as justification exists in other publicly deployed models and does not warrant the response — is buried in technical statements. The political response is compliance theatre.
The financial reality is dependency on the entity that just damaged its business.
The European Perspective
The Amazon-Anthropic-US-government triangle that unfolded between Thursday and Friday last week is the clearest possible illustration of the structural risk European enterprises face in their American AI dependencies. Every dimension of vendor risk that European AI procurement frameworks have considered — capability, safety, pricing, support — is now subordinate to a risk that has no equivalent in any current procurement standard. Call it political-financial alignment risk. Your AI vendor's commercial viability now depends not just on its technology, but on whether its competitors, investors and host platforms maintain stable, non-adversarial relationships with the US administration. None of these factors are within your control. None of them are disclosed in vendor contracts. None of them can be assessed in standard due diligence. The European Commission must respond. Not with a regulatory framework — which takes years — but with immediate procurement guidance for European public sector AI deployments and emergency funding for European AI alternatives. Mistral, Aleph Alpha and the European AI Factories programme are not theoretical sovereignty insurance. They are the only AI infrastructure that cannot be disabled by an American phone call. After last Friday, that distinction is no longer abstract. gafam.ai will be watching.
We are not first. We are right.
SOURCES
— TechTimes: Amazon Triggered Claude Fable 5 Shutdown: Investor, Cloud Host, Now Regulator
— TechTimes: Claude Accuracy Degradation After Fable Ban Has a Name: The Alignment Tax
— BigGo Finance: US Government Orders Anthropic to Shut Down Claude Fable 5 and Mythos 5 Worldwide
— MarkTechPost: Anthropic Disables Claude Fable 5 and Mythos 5 After US Government Order
— Yellow.com: Anthropic Shuts Claude Fable 5 And Mythos 5 After US Order