Three days after Anthropic launched Fable 5, billing it as the most capable model in the company’s history, the model was gone. Not restricted in one country or for one type of user, but pulled from public access entirely, worldwide. The trigger, according to the Wall Street Journal, traces back to a phone call from one of Anthropic’s own investors.
Amazon CEO Andy Jassy told Treasury Secretary Scott Bessent and other officials that Amazon researchers had used Fable 5 to extract information that could be used in cyberattacks, according to TechCrunch. The Commerce Department subsequently sent Anthropic CEO Dario Amodei a letter on June 12 invoking national security to bar foreign nationals, including Anthropic’s own foreign-born employees, from accessing Fable 5 and Mythos 5.
Why Amazon’s warning carries unusual weight
What makes this episode unusual is who raised the alarm. Amazon has invested roughly $13 billion in Anthropic and secured a $100 billion Amazon Web Services (AWS) infrastructure spending commitment from the company in return. Anthropic had been expected to pursue an IPO later in 2026.
More AI:
- Micron sits at the center of a red-hot chip rally
- IBM CEO sends blunt message on AI and quantum computing
- Anthropic CEO makes shocking admission about AI
An Amazon spokesperson confirmed the broad shape of the story without detailing it. “It’s not uncommon for governments to seek our counsel on potential security risks,” the spokesperson said, declining to share details of the discussions, according to TechCrunch. The spokesperson also confirmed that AWS itself was affected by the model cutoff, since Anthropic’s models run on Amazon’s cloud infrastructure.
What Amazon’s researchers reportedly found
According to The Next Web, Amazon researchers used a series of prompts to get Fable 5 to provide information that was supposed to be off-limits, specifically related to cyberattack techniques. Katie Moussouris, CEO of security firm Luta Security, reviewed the underlying report shared by Anthropic and described the technique to the Washington Examiner as a way of bypassing Fable’s guardrails to access the underlying Mythos model’s broader capabilities.
David Sacks, the former White House AI czar who now co-chairs the President’s Council of Advisors on Science and Technology, offered his own account in a post on X. He wrote that “a highly credible trusted partner of both Anthropic and the USG who was testing Fable came forward with a jailbreak of those guardrails,” and that the administration asked Anthropic CEO Dario Amodei to fix the issue or remove the model. “Dario refused,” Sacks added.
Yeh/Getty Images
Anthropic’s response and its scale
Anthropic confirmed it was complying while disputing the government’s framing. “We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users,” the company said, according to the Washington Examiner.
“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.”
The shutdown affected enterprise customers including banks and government agencies that had been using Mythos-class models for vulnerability discovery and complex reasoning tasks. Anthropic concluded that attempting to selectively block foreign nationals in real time was not technically workable, and disabled both models globally instead.
What this episode reveals about the broader AI landscape:
- Fable 5 and Mythos 5 trace back to Claude Mythos Preview, a model Anthropic had initially restricted to launch partners including Amazon, Google, Apple, Nvidia, and CrowdStrike under Project Glasswing, a program designed to let those companies find and fix security flaws before any wider release, a structure Anthropic had hoped would prevent exactly this kind of incident.
- This is not Anthropic’s first clash with the federal government. The company is currently suing the administration in a separate case after the Pentagon blacklisted it for refusing to remove restrictions on military use of Claude, according to Reuters.
- The timing compounds the pressure on Anthropic. Fable 5 launched on June 9 as the company’s flagship model, was pulled globally on June 12, and the company is simultaneously navigating IPO preparations that this kind of global shutdown does not make easier.
- The precedent question is what industry watchers are now focused on. If a major cloud provider can trigger an export control action against its own portfolio company by raising concerns directly with the Treasury Secretary, that gives large cloud and infrastructure partners a new form of leverage over the AI labs that depend on them.
What it means for the Amazon and Anthropic relationship
The practical relationship between the two companies is unlikely to change immediately. AWS remains central to Anthropic’s infrastructure, and Amazon’s $13 billion investment and $100 billion compute commitment are not going anywhere over a single incident. But the episode adds a new dimension to how that relationship is understood from the outside.
Amazon is simultaneously Anthropic’s largest financial backer, its primary infrastructure provider, and, in this case, the source of the security concern that led to a government-ordered shutdown of Anthropic’s flagship product. For an industry that has spent the past two years arguing that AI companies and their major partners are aligned on safety, this is the clearest public example yet of what happens when that alignment breaks down in practice, even temporarily, and even between two companies that need each other to succeed.
Related: The White House sends a shocking message to Anthropic
