Amazon CEO May Have Triggered the Global Shutdown of Anthropic's Newest AI

Anthropic's abrupt shutdown of its flagship Fable 5 and Mythos 5 AI models was preceded by warnings raised by Amazon CEO Andy Jassy to senior Trump administration officials, according to a report from The Wall Street Journal.



The report offers the clearest explanation yet for how the U.S. government arrived at its extraordinary decision to restrict foreign access to Anthropic's most advanced models, a move that ultimately forced the company to disable them for all users worldwide.


The global suspension carries immediate implications for the Apple ecosystem. Just days before the ban, Apple integrated Anthropic's technology directly into Xcode 27 to power autonomous coding agents for macOS and iOS developers. Apple also recently joined Anthropic's Project Glasswing, an initiative that grants select organizations early access to advanced models for identifying software vulnerabilities and strengthening critical infrastructure. Separately, researchers recently demonstrated how Anthropic's AI could be used to bypass advanced macOS security protections, highlighting why increasingly capable cybersecurity-focused models have drawn heightened scrutiny.

According to people familiar with the matter, Jassy told administration officials, including Treasury Secretary Scott Bessent, that Amazon researchers had found a sequence of prompts that caused Fable 5 to provide information that could potentially aid cyberattacks. The researchers reportedly discovered a way to elicit vulnerability-related information despite the model's safeguards.

White House officials quickly convened to assess the claims. Security researchers subsequently tested Amazon's findings, and officials concluded that restricting access by foreign governments, companies, and individuals was the most direct way to address the perceived risk. President Donald Trump later approved the action on national security grounds.

The revelations add new context to Friday's export-control directive that forced Anthropic to suspend access to Fable 5 and Mythos 5 globally. As previously reported, Anthropic said the government provided little detail about its concerns beyond what the company understood to be a potential jailbreak technique that allowed the models to identify a small number of software vulnerabilities.


Anthropic has strongly disputed the severity of Amazon's findings. The startup maintains that the vulnerabilities identified through the reported jailbreak are relatively basic and can also be discovered by other publicly available frontier AI models, including OpenAI's GPT-5.5. It argues that no universal jailbreak capable of broadly bypassing Fable 5's safeguards has been demonstrated.

Some cybersecurity researchers share parts of that assessment. Andrew Morris, founder of GreyNoise Intelligence, told the Journal the vulnerabilities reportedly uncovered by Amazon were still a long way from the sort of dangerous cybersecurity information that would typically raise major alarm. The report also notes there is no evidence Amazon's researchers were able to access Fable 5's ability to generate working exploit code, a capability protected by the model's security guardrails.

Amazon occupies a unique position in the dispute. The company is one of Anthropic's largest investors, supplies the AI startup with chips for its data centers, and deploys Anthropic's models internally to identify software vulnerabilities. An Amazon spokesperson declined to discuss specific conversations but noted that governments occasionally seek the company's perspective on potential security risks because of its role as a major cloud provider.

The report suggests that concerns about Anthropic extended beyond the specific findings shared by Amazon. According to people familiar with the administration's thinking, some officials had long questioned whether the company could adequately manage the security risks posed by increasingly capable AI systems. A call between administration officials and Anthropic CEO Dario Amodei on Friday reportedly reinforced those concerns.

The episode marks a significant escalation in federal oversight of advanced AI models. The Journal notes that the suspension follows broader administration efforts to increase scrutiny of frontier AI systems, including a recent executive order giving security officials greater oversight of AI models and discussions about a potentially larger government role in the industry.

Anthropic, which was founded in 2021 by former OpenAI researchers and has become one of the leading developers of coding-focused AI systems, continues to comply with the government's directive while working to restore access to Fable 5 and Mythos 5. The company has argued that recalling a commercially deployed frontier model over what it describes as a narrow jailbreak could effectively halt future frontier AI deployments across the industry.

Please download the iClarified app or follow iClarified on X, Facebook, YouTube, and RSS for further updates.