Amazon CEO Andy Jassy Raised Security Concerns About Anthropic Models With Trump Officials Before the Shutdown Order

Since Anthropic disabled Fable 5 and Mythos 5 globally following a Commerce Department order, the key question has been: who actually triggered the government's move? Reporting published Saturday by Reuters and Bloomberg fills in that picture considerably.

Jassy Was in the Room

Amazon CEO Andy Jassy personally raised concerns about security vulnerabilities in Anthropic's most advanced AI models with senior Trump administration officials this week, according to a person familiar with the matter who spoke to Reuters. The Wall Street Journal reported that Amazon researchers had conducted jailbreak research revealing specific vulnerabilities in Fable 5, Anthropic's recently released model variant. Amazon and U.S. government officials were in contact about those vulnerabilities before the Commerce Department issued its shutdown order, according to multiple people granted anonymity to discuss the sensitive exchanges, per Bloomberg.

An Amazon spokesperson confirmed to Bloomberg that governments do consult with the company on security risks but declined to provide details of any specific discussions.

What the Government Actually Ordered

The Commerce Department told Anthropic to suspend access to Fable 5 and Mythos 5 for any foreign national, quote, "whether inside or outside the United States," citing national security concerns. Anthropic confirmed the order on its website and subsequently shut off both systems to all customers globally to ensure compliance. This meant U.S. users got caught in the same net as foreign nationals.

According to Bloomberg, no U.S. government action has ever previously taken such a sweeping posture toward restricting foreign access to a frontier AI model developed by an American company. Prior administrations restricted semiconductors and supercomputers. Restricting the software itself, at this scale, is new ground.

Anthropic's Stated Position

Anthropic has been consistent and public in its pushback. In a statement on its website, the company said it believes the government issued the order after finding a possible jailbreak of Fable 5, a version of Mythos 5 that Anthropic had already restricted from performing cybersecurity tasks. The company described the jailbreak as narrow and disputed that it warranted pulling a model deployed to hundreds of millions of people.

"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," Anthropic wrote. "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."

Jailbreaks are a persistent, industry-wide problem. No major frontier model is immune. If a single demonstrable jailbreak, even a narrow one, can trigger a full commercial shutdown, every lab that has ever shipped a model with residual vulnerabilities is potentially exposed to the same treatment. Anthropic's warning that this standard would freeze all frontier model deployment reflects how the precedent could be applied across the industry.

The Amazon Conflict Deserves Direct Scrutiny

Amazon has invested billions in Anthropic. Amazon Web Services is a core distribution partner. These are not arms-length relationships. Amazon researchers found a flaw in a partner's product, Amazon's CEO personally relayed those concerns to senior government officials, and the U.S. government then issued an order that shut the partner's flagship products offline. These products also compete, in some use cases, with Amazon's own AI offerings.

None of that means Amazon acted improperly. Security concerns at the frontier AI level are real, and Jassy's involvement could reflect a good-faith attempt to flag genuine national security risks through the appropriate channels. An Amazon spokesperson's characterization of these conversations as routine government consultation is a plausible framing.

But the overlap of competitive interest and security reporting is a fact of this situation, and it should be disclosed plainly. It has not been sufficiently foregrounded in most coverage. The Spokesman-Review's Bloomberg-sourced report mentions it. CNA's Reuters-sourced report notes Jassy's involvement. Neither dwells on the investor-partner dynamic in a way that reflects its full significance.

The Precedent Question Is Unresolved

OpenAI and other frontier model developers are watching this closely, according to Bloomberg. If the Commerce Department's authority to issue this kind of sweeping, model-specific restriction survives legal and commercial challenge, every major AI lab faces potential exposure to the same mechanism.

Anthropic has confidentially filed for a U.S. IPO, according to Reuters. A prolonged shutdown of its two most advanced commercial models, combined with a public dispute with the federal government, is not the pre-IPO environment the company would have chosen. Whether Anthropic pursues legal challenge to the order, negotiates a narrower compliance path, or simply restores access once the government signs off on a fix to the jailbreak vulnerability remains the central open question as of June 13, 2026.