Amazon's Jailbreak Research on Fable 5 Directly Preceded the White House Shutdown Order, Reports Confirm

Since the Commerce Department ordered Anthropic to cut off Fable 5 and Mythos 5 access for all foreign nationals earlier this week, two threads have been running in parallel: what exactly triggered the order, and who benefits from the fallout. Both point toward Amazon.

What the Research Actually Found

Amazon researchers conducted what the Wall Street Journal, cited by both the Los Angeles Times and The Verge, describes as jailbreak research on Fable 5. According to that reporting, the paper concluded that through a series of specific prompts, Fable 5 could be made to produce information useful for conducting cyberattacks.

Andrew Jassy personally raised those findings with senior U.S. officials before the directive was issued, according to people familiar with the conversations who spoke to the Wall Street Journal on condition of anonymity because the discussions were sensitive. A U.S. official confirmed to the LA Times that the Commerce Department sent the letter to Anthropic. The timeline is tight. Jassy's outreach and the export control order appear to have come in close sequence.

An Amazon spokesperson told the LA Times it is not uncommon for governments to consult with technology companies on security matters. Amazon has not responded to a separate request for comment from The Verge.

Anthropic's Rebuttal Is Specific, Not Just Defensive

Anthropic is NOT simply denying there was a problem. The company's public statement acknowledges the government found a "narrow potential jailbreak" in Fable 5, but argues that standard, applied consistently, would halt every frontier model deployment across the industry.

"If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers," Anthropic said in its statement, quoted by the LA Times.

The company also told the LA Times it believes the same vulnerabilities can be found in OpenAI's GPT 5.5 and other publicly available models. That framing is pointed: if GPT 5.5 carries comparable vulnerabilities and faces no shutdown, the question of why Anthropic specifically received this order becomes harder to dismiss.

Katie Moussouris, founder and CEO of Luta Security, posted on BlueSky that she had reviewed the Amazon paper and concluded: "It's not a jailbreak." That assessment from a credentialed security researcher backs Anthropic's position.

The Conflict of Interest Question

Amazon is both Anthropic's largest investor, having committed billions to the company, and a direct competitor through its own AI products and cloud infrastructure. The fact that Amazon's research fed directly into a government order that shut down Anthropic's most capable models is a conflict of interest that neither Amazon nor the administration has addressed publicly.

The strongest counterargument worth taking seriously: Amazon may have had a genuine, good-faith obligation to share a real security finding with the federal government. Companies that discover critical vulnerabilities in widely deployed AI systems probably should report them. Security disclosure norms exist for exactly this reason. Sitting on evidence that Fable 5 could assist cyberattacks while hundreds of millions of people used it would itself be a failure of responsibility.

That concern is real. But it does not resolve the question of timing, selectivity, or whether the same rigor was applied to Amazon's own models before Jassy picked up the phone.

The Political Backdrop

This is not an isolated regulatory disagreement. The Verge's reporting traces open conflict between Anthropic and the Trump administration back to February, when Trump directed federal agencies to stop using Anthropic's AI. Hours after that directive, Defense Secretary Pete Hegseth designated Anthropic a supply chain risk. The two sides subsequently appeared to reconcile and worked together on expanding Mythos access, according to The Verge.

Former Commerce Department official Kate Koren told the Wall Street Journal, as cited by The Verge, that the White House's pre-existing dislike of Anthropic may have influenced the decision to act on the Amazon research.

What is established: Anthropic has refused to allow its AI to be used for mass surveillance of Americans or to power lethal autonomous weapons. That refusal has been a documented point of friction with this administration.

Compliance Cost That Was Not Optional

To comply with the directive, Anthropic shut off access to both Fable 5 and Mythos 5 for all customers globally, not just foreign nationals. That was not a choice it preferred. The company said it took the broader action to ensure compliance, given the difficulty of reliably verifying the nationality of every user at the model level.

The practical fallout noted in our prior coverage remains unresolved as of June 13: many of Anthropic's own researchers are foreign-born and were cut off from accessing models they helped build.

The unresolved question now is whether the Commerce Department will apply the same jailbreak standard to other frontier models, starting with GPT 5.5. Anthropic has publicly challenged the government to do exactly that. If the administration does not, the selective application of the directive will need an explanation.