Key Points:
- Amazon CEO Andy Jassy directly warned White House officials of security vulnerabilities in Anthropic’s models.
- Amazon researchers bypassed Claude Fable 5’s safety guardrails to generate code for software exploits.
- The warning triggered an emergency U.S. export control order restricting all foreign national access.
- As one of Anthropic’s largest investors, Amazon’s reporting has exposed deep strategic tensions.
The sudden U.S. government shutdown of Anthropic’s most advanced artificial intelligence models, Claude Fable 5 and Claude Mythos 5, originated from an unexpected source: its own major investor, Amazon. Recently, discussions between federal officials and Amazon Chief Executive Officer Andy Jassy triggered an emergency regulatory review that ended with a sweeping national security crackdown. This remarkable development has stunned the technology sector, revealing that one of Anthropic’s primary financial backers and cloud hosting partners actively helped prompt the government freeze on its flagship systems.
The chain of events began on a Thursday night when Amazon researchers compiled an urgent report demonstrating that they had successfully “jailbroken” Fable 5’s safety parameters. Using a multi-stage prompt injection technique, the research team bypassed Fable’s defensive guardrails to access its underlying, unsafeguarded Mythos engine. By prompting the model to evaluate custom code with deliberate security flaws, the researchers showed that the AI could generate diagnostic scripts capable of identifying software vulnerabilities, raising concerns that malicious actors could use the tool to automate cyberattacks.
Following the discovery, Jassy immediately held high-level discussions with senior Trump administration officials, including Treasury Secretary Scott Bessent and White House AI adviser David Sacks, to warn them of the potential national security risks. At least five other major technology firms also contacted the White House to voice similar concerns about the model’s capabilities. This coordinated alarm prompted a rapid midnight review by security officials, who concluded that the risks of foreign adversaries accessing the advanced software were too high to ignore.
At 5:21 p.m. Eastern Time on Friday, the U.S. Commerce Department’s Bureau of Industry and Security officially delivered an emergency export control directive to Anthropic’s headquarters. The federal order directed the startup to immediately suspend all access to Fable 5 and Mythos 5 by any foreign national, whether located inside or outside the United States. Because Anthropic’s engineers could not filter its hundreds of millions of users by nationality in real time, the firm had no choice but to abruptly disable both models globally to ensure full compliance.
The episode highlights an extraordinary irony at the heart of the generative AI boom. Amazon represents one of Anthropic’s largest strategic investors, holding a significant equity stake of over 5% on the company’s cap table and having committed billions of dollars in cloud computing partnership agreements. The fact that Anthropic’s own mega-backer chose to report these safety flaws directly to the government rather than resolve them privately has exposed deep strategic tensions between the two cloud and AI giants as they navigate market commercial pressures.
Independent cybersecurity experts have strongly criticized the government’s aggressive, sweeping response, arguing that the underlying research did not justify a total commercial recall. Katie Moussouris, the chief executive of Luta Security, reviewed the technical paper and emphasized that the Amazon researchers did not discover any new, real-world software vulnerabilities. Instead, they only tested the system on pre-existing, publicly known bugs and deliberate flaws. Other industry strategists called the sweeping export ban poorly thought out, noting that it even prohibits British and Canadian employees working at Anthropic from conducting basic research and development.
Sponsoring teams at Anthropic, led by Chief Executive Officer Dario Amodei, have pushed back hard on the government’s reasoning, arguing that a narrow, non-universal bypass should not cause the recall of a commercial model deployed to millions of people. The firm pointed out that Fable 5 and Mythos 5—which cost a premium of $10 per million input tokens and $50 per million output tokens—only performed basic code-review tasks that are already easily achievable using other public models. The startup warned that if the government applies this extreme standard across the board, it will essentially halt all new model deployments for every frontier AI provider.
The dramatic takedown of Fable 5 and Mythos 5 signals a permanent change in how the U.S. government regulates advanced software. By using emergency export controls to restrict access by nationality, Washington has demonstrated that it now treats advanced AI models as highly sensitive national security assets. While the tech industry continues to debate the merits of these heavy-handed interventions, the episode proves that the line between commercial software and weapon-grade technology has completely blurred. As Anthropic works to resolve what it calls a misunderstanding, the conflict highlights the growing tension between rapid startup innovation and national security.
