Fable 5 and the New Reality: If It’s Powerful, It’s Political -

The Real Trigger: Amazon Did It

The most explosive new detail: it wasn’t a random hacker or internet jailbreaker who triggered the government order — it was Amazon, Anthropic’s own largest investor.

According to the Wall Street Journal, corroborated by Reuters and The Information, Amazon researchers discovered that Fable 5 could be prompted to surface security vulnerabilities in specific software codebases, and Amazon CEO Andy Jassy personally escalated the concern to Treasury Secretary Scott Bessent . National Cyber Director Sean Cairncross and Commerce Secretary Lutnick were both pulled into the conversations. David Sacks, the White House AI czar, described the source only as a “highly credible trusted partner” .

The conflict of interest here is extraordinary: Amazon holds a cumulative ~$13 billion stake in Anthropic, operates a $100 billion AWS cloud spending commitment running the other way, hosts Fable 5 via Amazon Bedrock, and holds a board seat . One of the companies most commercially entangled with Anthropic helped prompt the government action that knocked Anthropic’s flagship launch offline, eleven days after Anthropic had confidentially filed for an IPO.

Was Amazon concerned about legal liability for hosting a jailbroken model? Nobody is saying it publicly. Amazon told reporters only that it is “not uncommon for governments to seek our counsel” on security risks .

The “Jailbreak” That Wasn’t (According to Experts)

The only named independent expert who actually read the Amazon report is Katie Moussouris — the Luta Security CEO who built Microsoft’s bug-bounty program and designed the Pentagon’s first. Her verdict after reviewing the findings at Anthropic’s request:

“Not a jailbreak, Defense Oriented Prompting (DOP), capabilities defenders need. If national defense was the goal, this just scored an own goal against us.”

Chris McGuire of the Council on Foreign Relations called the across-the-board restriction “highly questionable” . Anthropic itself notes the jailbreak was essentially asking the model to read a codebase and flag software bugs — the same result achievable from GPT-5.5 and other publicly available models that remain fully online.

Thousands of hours of pre-launch red-teaming by the U.S. government, the U.K. AI Security Institute, and outside groups found no universal jailbreak before launch.

The Government’s Counter-Narrative

The administration isn’t backing down. David Sacks posted publicly that a trusted partner found a working jailbreak; the administration asked Dario Amodei to fix it or pull the model, and “Dario allegedly refused” . Sacks’ position: Anthropic spent months warning the world that Mythos-class models were uniquely dangerous; so a bypass that exposes those capabilities is precisely the kind of thing that justifies intervention.

Defense Secretary Pete Hegseth piled on with a “told you so,” writing:
“Three months ago, @DeptofWare kicked @AnthropicAI out of our building, forever. Every passing day proves why that was the right move” .

A darker claim — that the trigger involved access from China — rests on a single Semafor source and is disputed by Anthropic, which says it blocks all access from inside China and the issue was never raised with them .

The Deeper Political Backstory

This showdown didn’t start on June 12. The Pentagon designated Anthropic a “supply chain risk” earlier this year after Anthropic insisted on contract language barring its AI from domestic mass surveillance and fully autonomous weapons systems. The DOD would only accept restrictions to “all lawful use.” Negotiations broke down. Anthropic sued the Trump administration over the designation.

That lawsuit is still unresolved. Now Anthropic is simultaneously fighting a Pentagon blacklist and a Commerce Department export control, two separate legal fights with two separate agencies of the same government.

Sacks has denied the Fable action is retaliation, and there is no public evidence it is. But Hegseth’s gleeful “told ya so” post isn’t helping that case.

What’s Happening Right Now (June 15)

As of this morning:

Fable 5 and Mythos 5 remain offline; models are still down with no restoration date set
Senior Anthropic technical staff have been dispatched to Washington D.C. to negotiate directly with the administration
Sources on both sides told Axios they are “eager to resolve it” — but the two parties are telling fundamentally different stories about what happened
The Trump administration is reportedly unsatisfied with Anthropic’s handling of the jailbreak complaint, saying Anthropic has not “engaged in a serious manner”
Anthropic is calling it a misunderstanding and working toward restoration

The Five Big Implications

1. The IPO Is Now Complicated

Anthropic confidentially filed for an IPO eleven days before this happened. Barron’s noted the timing directly: on the same day SpaceX pulled off the largest IPO in history, Anthropic’s flagship model went dark. Investors are now asking a hard question: can a frontier AI company command a premium valuation if its most powerful products face sudden, unpredictable restrictions on global distribution?

2. Allied Nations Were Cut Off Too

The export control didn’t just block adversaries. It swept in NATO, the EU’s cybersecurity agency ENISA, Canadian and Indian users, the Korea Internet & Security Agency, SK Telecom, and Samsung, all of which were vetted Glasswing partners . The European Commission said emergency measures should not discriminate against partners. French officials reached for the language of “technological sovereignty”. The subtext — that AI infrastructure controlled in Washington can be switched off in Washington — is now being said aloud by America’s allies.

3. The “Deemed Export” Rule Changes Everything

The legal mechanism the government used, the decades-old “deemed export” rule, has never before been applied to a deployed commercial AI model. Applying it here means the government now has a legal blueprint to treat any frontier AI model as an export-controlled technology, the same category as weapons components and military hardware. Every AI lab is reading this carefully.

4. An Executive Order Just Made This the New Normal

Trump signed an executive order this month directing agencies to establish a mechanism for the government to get early access to powerful models before deployment. The Fable shutdown is described in the article as “what the involuntary version looks like” — meaning the voluntary pre-clearance framework is the carrot, and emergency export controls are the stick .

5. China and Open-Source Win Either Way

Enterprises and governments globally are now accelerating toward multi-provider routing, local model fallbacks, and open-weight models; precisely the direction Chinese open-source labs are now aggressively marketing. The restriction that was meant to protect the American AI advantage may end up pushing allies and enterprises toward alternatives that Washington has no control over at all.

The Irony Anthropic Can’t Escape

Anthropic built its brand on being the “responsible” AI company, the one that welcomed government oversight and argued publicly that regulators should have the power to stop unsafe deployments. Now it’s receiving exactly that, through a process that is, by Anthropic’s own stated standards, not “transparent, fair, clear, and grounded in technical facts”.

Its objection isn’t that no model should ever be stopped. It’s that this is the wrong way to stop one, an emergency directive with no public record, no disclosed evidence, no appeal process, and a 90-minute compliance window. That’s a harder argument for a company that spent years naming the danger and marketing the restraint. 🎭

View this post on Instagram

Teams, Tech, and Training