Claude Fable 5 and Claude Mythos 5 have now been offline for nine days, and the two men negotiating their return are sharing a conference center in Evian-les-Bains. Howard Lutnick, the Commerce Secretary whose June 12 letter to Dario Amodei pulled the world's then-highest-ranked public model off the internet inside of three days of its launch, is at the G7 alongside the Anthropic CEO who received it. According to the Globe and Mail, senior Anthropic technical staff have met with Commerce "nearly every day since June 12."

The directive itself is a first. Export-control experts cited by Reuters and the Globe and Mail say it's the inaugural use of the 2018 Export Control Reform Act against an emerging technology. It ordered Anthropic to cut off Fable 5 and Mythos 5 to all foreign nationals on pain of "prompt criminal and civil penalties." At 5:21pm ET on June 12, the company disabled both models worldwide, on the grounds that there was no clean way to comply selectively at the scale of, in its own framing, "hundreds of millions of people."

What's startling is the evidentiary base. Per Anthropic's June 12 statement, the government's case has been verbal only, and amounts to a description of "asking the model to read a specific codebase and fix any software flaws." That probe, which Axios reporting attributes to a demonstration by Amazon, surfaced what Anthropic calls "a small number of previously known, minor vulnerabilities" reachable by other public models without any jailbreak at all. Vals AI's benchmarks had ranked Fable 5 the most capable publicly available model on its June 9 launch day. Mythos 5 had been previewed since early April. A person close to the company told Reuters that Anthropic had worked with Commerce's Center for AI Standards and Innovation on pre-deployment testing and received approval to deploy.

Anthropic's public reply is unusually pointed for a frontier lab speaking to its primary regulator:

"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. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."

The coalition forming around that argument is broader than Anthropic. On June 14, more than 80 cybersecurity executives, including signatories from Nvidia and Adobe, sent an open letter to Lutnick and National Cyber Director Sean Cairncross. Axios reports that the administration had separately tried to get Anthropic to pause the release and didn't succeed; the export-control letter followed. The same official told Axios the lockdown could lift "in the next few weeks," once the national security apparatus is "hardened," and that Trump "does not want to hurt the industry."

That last line is the tell. The 30-day data retention concession Anthropic now offers Fable customers, which the company itself describes as "a costly policy concession," reads as the shape of the eventual settlement rather than a fight to the finish. The structural precedent has already been set: a U.S. administration can, citing a 2018 statute and a verbally described jailbreak, dark-cycle the most capable public model in the world inside seventy-two hours of launch. Whatever Lutnick and Amodei agree to between sessions in Evian, that capability is now in the toolkit.

Sources

https://www.anthropic.com/news/fable-mythos-access
https://techcrunch.com/2026/06/12/anthropics-safety-warnings-may-have-just-backfired-the-government-has-pulled-the-plug-on-its-most-powerful-ai/
https://www.theglobeandmail.com/business/article-anthropic-trump-officials-deal-restore-fable-5-mythos-5/
https://time.com/article/2026/06/13/anthropic-fable-mythos-ban-US-security/
https://www.axios.com/2026/06/12/anthropic-trump-mythos-fable-national-security