A Wall Street Journal report reveals that Amazon CEO Andy Jassy alerted Treasury Secretary Scott Bessent about a jailbreak of Anthropic's Fable 5 AI, leading to a global ban and raising questions about corporate conflicts of interest in AI regulation.
How a Late-Night Phone Call Derailed Anthropic ’s Flagship AI Launch
Just four days after Anthropic unveiled its most advanced AI models— Fable 5 and Mythos 5 —a late-night call from Amazon CEO Andy Jassy to Treasury Secretary Scott Bessent triggered a cascade of government actions that forced the company to yank both models worldwide. According to a detailed report inThe Wall Street Journal, Jassy alerted Bessent and other senior officials that Amazon researchers had successfully jailbroken Fable 5 , extracting sensitive cybersecurity information the model’s safeguards were designed to block.How a Late-Night Phone Call Derailed Anthropic’s Flagship AI Launch
Just four days after Anthropic unveiled its most advanced AI models—Fable 5 and Mythos 5—a late-night call from Amazon CEO Andy Jassy to Treasury Secretary Scott Bessent triggered a cascade of government actions that forced the company to yank both models worldwide. According to a detailed report inThe Wall Street Journal, Jassy alerted Bessent and other senior officials that Amazon researchers had successfully jailbroken Fable 5, extracting sensitive cybersecurity information the model’s safeguards were designed to block.
The U.S. government responded with unusual speed, imposing export controls that barred foreign nationals and users outside the United States from accessing the models. Anthropic, citing an inability to filter foreign users in real time, complied by shutting down access for all customers globally—including its own researchers, many of whom are foreign-born. The move marked one of the first times export controls have been used to block a commercial AI product based on security concerns, setting a precedent that could reshape the industry.
The Jailbreak: What Amazon Researchers Uncovered
Amazon’s security team used carefully crafted prompts to bypass Fable 5’s guardrails, revealing vulnerabilities in at least four software programs. The technique, which Anthropic later characterized as a “narrow potential jailbreak,” exposed what the company called “a small number of previously known, minor vulnerabilities.” But for U.S. officials, the implications were severe enough to invoke rare export controls under the International Emergency Economic Powers Act.
“We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users,” Anthropic said in a statement. “However, 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 CEO Dario Amodei pushed back further, arguing that the same jailbreak technique works on other models, including OpenAI’s GPT-5.5. “The capabilities causing concern are already available in other publicly accessible models,” the company stated, warning that applying this standard industry-wide would effectively halt AI development.
A Conflict of Interest: Amazon as Investor, Partner, and Whistleblower
The revelation adds an uncomfortable dimension to Amazon’s role. The tech giant is one of Anthropic’s largest investors, having poured billions into the company and secured a $100 billion cloud spending commitment in return. Amazon’s AWS is also the cloud platform hosting Anthropic’s models—meaning the shutdown directly affected its own infrastructure. Critics argue this creates a conflict of interest: the same company that bankrolls Anthropic’s operations also flagged its models as dangerous to the government.
“It’s an ouroboros of corporate dependency and regulatory blowback,” noted one industry analyst. Amazon, however, defended its actions. A spokesperson told the WSJ that while it’s “not uncommon for governments to seek our counsel on potential security risks,” the company does not “share the details of those discussions.” The episode has sparked debate about whether Amazon’s move was a responsible safety measure or a strategic check on a partner that is also a potential rival in the AI race.
Broader Implications for AI Regulation and National Security
The episode marks a significant escalation in the U.S. government’s approach to AI safety. It is one of the first times export controls have been used to block access to commercial AI models based on security concerns, signaling that Washington is willing to take aggressive action to prevent AI-powered cyberattacks. The decision also highlights the growing role of corporate whistleblowing in shaping AI regulation—a trend that could intensify as AI models become more powerful.
For Anthropic, the blow is both financial and reputational. Fable 5 and Mythos 5 were its flagship products, and their abrupt withdrawal undermines customer trust. The company is reportedly working on fixes to the jailbreak and may seek to re-release the models under stricter controls. Meanwhile, the incident has opened a technical rift in the AI community: some experts side with Anthropic, warning that overreaction could stifle innovation and cede ground to foreign competitors, while others argue that the potential for misuse—especially in cybersecurity—justifies strict controls.
What’s Next?
The White House has not commented on the specific interactions with Amazon, but officials have signaled that AI security will remain a top priority. The Treasury Department and Commerce Department are expected to review export control policies for AI models, potentially creating a new regulatory framework. Anthropic is lobbying for a more nuanced approach, arguing that the vulnerability is not unique and that a blanket ban sets a dangerous precedent.
- Key players:Amazon (investor, whistleblower, cloud provider), Anthropic (developer, victim), U.S. government (regulator).
- Timeline:Models launched Monday; jailbreak discovered by Amazon researchers; Jassy called Bessent late Thursday; government imposed controls Friday; Anthropic shut down models Friday.
- Impact:Global customers locked out; foreign researchers at Anthropic affected; industry-wide trust shaken; potential regulatory precedent set.
As the dust settles, one thing is clear: the line between partner and watchdog in the AI industry has never been thinner—and the stakes have never been higher.
Please follow us on Google to support us Anthropic Fable 5 Mythos 5 Andy Jassy AI jailbreak export controls US government AI security cyberattack Claude Trump administration Scott Bessent Dario Amodei AWS