How Amazon CEO Andy Jassy's Warning Triggered the U.S. Ban on Anthropic's Most Powerful AI

According to the report, Amazon researchers ran a series of prompts on Anthropic’s newly released Fable 5 and successfully extracted information that could be used to aid cyberattacks—material the model’s safeguards were explicitly designed to keep off-limits . This wasn't just a superficial glitch; sources characterized it as a genuine safety bypass that exposed potentially dangerous capabilities .

Jassy’s decision to take these concerns directly to the highest levels of the administration set the stage for a dramatic federal intervention. As an investor in Anthropic, Amazon’s move was seen by some as a responsible act of red-teaming and by others as a complex, perhaps competitive, maneuver from a tech giant that both backs its partner and fiercely competes with it .

The Government’s Unprecedented Hammer: An Export Control Ban

Jassy’s conversation didn't just spark a review; it ignited a rapid-fire export control action. The Trump administration determined that Anthropic’s models posed a sufficient national security risk to warrant immediate export restrictions. On the evening of Friday, June 12, Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei formally placing Fable 5 and Mythos 5 under export limitations .

The directive was sweeping. It suspended all access to the models by any foreign national, whether inside or outside the United States—a definition that even included Anthropic’s own foreign-born employees . This placed the company in an impossible position.

Because Anthropic could not reliably distinguish foreign from domestic users in real time to comply with the narrow legal order, it made a drastic choice. To ensure full compliance, it abruptly disabled both Fable 5 and Mythos 5 for all users worldwide at exactly 5:21 p.m. ET . Models that had launched just 72 hours earlier on June 9 went dark for everyone .

A Clash of Narratives: How Amazon, Anthropic, and David Sacks Responded

What followed was a fierce public dispute over the severity of the threat and the appropriateness of the government’s response.

Amazon’s Strategic Silence

Amazon did not publicly deny the WSJ’s report. While the company issued no formal statement directly addressing Jassy’s conversations, its position, cited by sources, was that it was acting responsibly by alerting officials to a genuine vulnerability it had discovered during internal security research . This framing presented Amazon as a good-faith security researcher, though its role as Anthropic's major investor added a layer of complexity to the optics.

Anthropic’s Pushback: Minor Flaws, Not a Catastrophe

Anthropic immediately contested both the substance and the scale of the government’s action. In a public statement, the company confirmed the directive but called it a "misunderstanding" . It acknowledged reviewing the specific jailbreak technique but characterized the findings as extremely minor. Anthropic stated the method only identified "a small number of previously known, minor vulnerabilities" and argued that other publicly available models, including OpenAI's GPT-5.5, were capable of finding similar flaws .

The company pushed back hard against the precedent, stating, "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" .

David Sacks Alleges Refusal to Remediate

The government’s narrative, however, was starkly different. Former White House AI and crypto czar David Sacks, now a co-chair of the President's Council of Advisors on Science and Technology, offered a detailed public account that shifted the blame squarely onto Anthropic’s executives.

Sacks posted that a "highly credible trusted partner of both Anthropic and the U.S. government"—widely assumed to be Amazon—uncovered a jailbreak capable of stripping away Fable 5’s guardrails to expose the full, more dangerous capabilities of the Mythos model beneath . According to Sacks, the administration then presented Anthropic CEO Dario Amodei with a clear choice: either fix the vulnerability or de-deploy the model .

Sacks claimed that Amodei refused both options. "The Admin asked Dario to fix the jailbreak or de-deploy the model. Dario refused," Sacks stated, adding that the export control was a necessary and "reluctantly" applied measure to manage a risk that the company would not address . Sacks further accused Anthropic of bad faith, criticizing the company for previously hyping Mythos as a cybersecurity superweapon needing regulation, only to dismiss concerns when those same capabilities were shown to be at risk of leak .

This clash of accounts—a company claiming victimization by a government overreaction to a minor bug, versus a top official alleging a refusal to fix a genuine weapons-grade risk—defines the most significant AI policy showdown to date. The White House maintained the ban was a temporary measure that could be lifted if Anthropic remediated the security flaw .