Anthropic Disagrees With Its Own Shutdown of Claude Fable 5, And That’s the Real Story
Three days after Anthropic launched Fable 5 to the public, the US Commerce Department ordered the company to cut off all foreign nationals from the model, but the company's statement reveals it does not believe the jailbreak that triggered the order is actually dangerous.
Anthropic’s most advanced models are gone. But a more interesting tension underneath is that Anthropic does not agree with the government’s reasoning, said so in writing, and shut the models down anyway.
The US government issued a directive on June 12, 2026, to suspend all access to the newly launched Fable 5 and Mythos 5 by any foreign national, whether inside or outside the US, including foreign national Anthropic employees.
The models had been live for exactly three days. The Fable 5 launch was framed as the company finally delivering on its Project Glasswing promise: public access to Mythos-class capability. That promise lasted 72 hours.
What Amazon Actually Found and Why It Matters
As Fortune reports, Amazon CEO Andy Jassy first raised concerns with senior administration officials after Amazon researchers used a series of prompts to make the Mythos-class model reveal restricted cyberattack information.
The outlet cites an unnamed source familiar with Amazon’s discussions, who says the government asked Amazon for feedback on the new Anthropic model.
That detail reframes the entire sequence: this was not Amazon stumbling onto a flaw; it appears the White House proactively asked one of its largest cloud partners to stress-test a rival’s frontier model, and Amazon delivered a result the government acted on within 24 hours.
David Sacks, Chair of the President’s Council of Advisors on Science and Technology, stated on X that the government was warned that Fable 5 could be jailbroken and claimed CEO Dario Amodei dismissed the risk and declined to fix it.
This arrives amid an active, months-long legal dispute between Anthropic and Washington over the Pentagon’s “supply chain risk” designation.
Anthropic’s Defence, In Its Own Words
Upon reviewing a demonstration of the specific exploit, Anthropic countered that the technique merely identified a handful of minor, previously known vulnerabilities.
The company noted that these security flaws are relatively simple and easily discoverable by other publicly available frontier tools, including OpenAI’s ChatGPT, without Mythos-specific uplift.
By maintaining that its defense-in-depth strategy aligns the risks of Fable 5 with standard industry benchmarks, Anthropic states it had not received any reports of non-universal jailbreaks resulting in actual harm.
In plain terms, Anthropic disagrees with the basis for the order but is complying with it. It also said that governments should be able to block unsafe deployments through a transparent process.
Ultimately, it’s trying to walk a careful line, defending its safety approach while still acknowledging that regulatory override ultimately stands.
The China Angle Nobody’s Statement Mentions Directly
The export controls on Anthropic’s Mythos-class models were imposed partly amid U.S. government concerns that a China-linked group may have accessed the technology.
This ties back to a Bloomberg report in April describing an unauthorized breach via a third-party contractor’s environment shortly after the launch of Project Glasswing.
Senior Anthropic technical staff is reported to have traveled to Washington over the weekend to meet with officials.
The move is believed to be the first instance of the U.S. government using export controls to restrict access to a widely deployed commercial frontier AI model.
The outcome of ongoing discussions between Anthropic and the administration could set an important precedent for how future advanced models are regulated.
Source: Statement on the US government directive to suspend access to Fable 5 and Mythos 5
