Anthropic Confirms Fable 5 and Mythos 5 Are Still Offline. The Jailbreak That Triggered the Shutdown May Have Come From Amazon.

Since Commerce Secretary Howard Lutnick's export control directive landed late Friday, June 12, Fable 5 and Mythos 5 have remained offline for every customer worldwide, including Anthropic's own foreign national employees. That hasn't changed as of June 13, 2026.

What is new is where the triggering jailbreak report apparently came from.

The Amazon Connection

The Wall Street Journal reports that the Commerce Department was alarmed by a jailbreak report submitted by Amazon, one of Anthropic's largest investors and its primary cloud compute provider. If accurate, it means a commercial partner, not a government lab, handed regulators the evidence that forced a global shutdown.

Amazon has not publicly commented on this.

The order, as Anthropic described it on June 13 according to The Stack, required the company to "suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees." Because Anthropic cannot currently verify the citizenship of every user in real time, the company concluded it had no choice but to pull both models entirely.

What the Government Actually Has

Anthropic's core complaint about the process is specific and on the record. The company says it received only "verbal evidence of a potential narrow, non-universal jailbreak" with no written technical documentation or formal finding.

Anthropic reviewed a demonstration of the technique and concluded the vulnerabilities were "relatively simple" and already available through other publicly accessible models, including OpenAI's GPT-5.5, according to the company's published statement cited by the BBC and The Guardian.

There is also a separate, publicly documented jailbreak from the UK's AI Safety Institute. On June 9, Xander Davies, the Red Team lead at the AISI, posted on X that within hours of access he and colleague Giorgi Giglemiani had developed a jailbreak that "extracted malicious responses to single-turn question-answering," and extended it over two additional days to enable "multiple steps of malicious agentic tool-calls." The Stack reported this on June 13. Whether the government's order was triggered by the AISI work, the Amazon report, or both has not been confirmed.

The Strongest Case for the Government's Action

Critics of Anthropic's pushback have a fair point: Anthropic itself spent years warning the public that its frontier models were too powerful to release carelessly. Having built its brand around safety-first messaging, the company now finds itself arguing that a jailbreak report is overblown and that the resulting vulnerabilities are widely available from other models.

More substantively, the government's concern isn't frivolous. A model capable of systematically identifying software vulnerabilities, even through a narrow jailbreak, is a genuine counterintelligence risk if accessible to foreign adversaries. Export controls on AI outputs rather than just on AI chips represent a real policy frontier, and the government's instinct to get ahead of that has a rational basis.

The Legitimate Problem With How This Was Done

That said, Anthropic's procedural objection stands on its own merits regardless of the underlying policy question. The company received no written technical finding, no specific vulnerability documentation, and was given no formal review process before being ordered to disable a product deployed to hundreds of millions of users. Anthropic noted, with some credibility, that as recently as Wednesday, June 11, it had publicly called for greater government oversight of AI, including the authority to block models with unacceptable risks. Its objection isn't to oversight; it's to an order it describes as bypassing "fair and fact-based regulation."

Harvard Law Review editor Ben Murphy, quoted by The Stack, framed the broader implication: requiring proof of citizenship to access commercial software services "might previously have been unthinkable" and reflects a deepening balkanization of technology with no clear policy framework guiding it.

What Else Is in the Background

This shutdown doesn't happen in isolation. According to The Guardian, Anthropic's relationship with the Trump administration ruptured earlier this year after the company refused to allow the U.S. military to use its models for domestic surveillance and fully autonomous weapons systems. The government responded by placing Anthropic on a supply chain blacklist, set to take effect later this year. The Guardian also notes that a separate ongoing lawsuit between Anthropic and the administration over government agency access to its tools has been unresolved.

The BBC reported June 13 that it approached the U.S. Department of Commerce for comment. No response had been received as of the BBC's publication.

The Open Question

Anthropic says it is "scrambling to reverse" the order, according to The Stack. What remains genuinely unresolved is whether the Commerce Department will provide written technical evidence to support its directive and if it does, whether that evidence came from Amazon, the AISI, or somewhere else entirely. Until that documentation is on the table, both Anthropic's characterization of the order as a "misunderstanding" and the government's characterization of it as a legitimate national security action remain unverified.