Anthropic’s two most advanced AI models, Claude Fable 5 and Mythos 5, are finally back online after 19 days it was shut down. However, the bigger news remains what Anthropic had to give up to make that happen.
How the Shutdown Started
The trouble began just three days after Anthropic released Claude Fable 5 to the public on June 9, where the U.S. Department of Commerce ordered the company to cut off all access to foreign users, whether inside or outside the United States, citing national security concerns.
The order followed a threat intelligence report from Amazon’s researchers claiming to have jailbroken Fable’s cybersecurity capabilities. Anthropic complied immediately but pushed back on the characterization, later noting that other systems including ChatGPT 5.5, Claude Opus 4.8 and Kimi K2.7 could identify the same vulnerabilities the Amazon report had flagged.
The Deal That Brought the Models Back
On June 30th, Commerce Secretary Howard Lutnick announced on X that the Bureau of Industry and Security had withdrawn the controls, and a license is no longer required to export, reexport or transfer either model. However, that reversal came with conditions attached. Anthropic confirmed it had reached an agreement with the Commerce Department to deploy both models with new guardrails and classifiers meant to address jailbreaks.
Specifically, Anthropic said it trained a new safety classifier built to detect and block the exact technique described in the Amazon report. When that classifier catches a request, the user gets notified and the task is automatically rerouted to Claude Opus 4.8 instead. Anthropic said the new safeguards were stress tested by the federal Center for AI Standards and Innovation and will block the flagged behavior about 99.9% of the time.
Anthropic also acknowledged the tighter filtering will also catch some legitimate work. Routine coding, debugging and other benign cybersecurity tasks are expected to get flagged more often, meaning defenders doing normal security work may find Fable 5 harder to use than before the shutdown.
What Is Rolling Out and When
Fable 5 returned globally to users on Wednesday across Claude.ai and Claude Code. Access through Amazon Web Services, Google Cloud and Microsoft Foundry will follow, although Anthropic has not given a firm date for those platforms.
For Mythos 5, which is a more restricted version, it has already been restored for a small group of vetted U.S. organizations through Anthropic’s Project Glasswing cybersecurity program, following separate government approval granted on June 26. Anthropic says it plans to keep expanding that access to more partners over time.
A Bigger Pattern With Washington
This standoff between the U.S. government and a private company was not an isolated incident. Earlier this year, Anthropic clashed with the Pentagon over guardrails preventing military use of Claude models for autonomous weapons or mass surveillance, a dispute that led Defense Secretary Pete Hegseth to label the company a “supply chain risk.”
Christopher Padilla, a former Commerce Department export official in the George W. Bush administration, said the episode reflects an administration that has been inconsistent on AI policy, pointing out that the same agency restricting Anthropic’s models has taken a permissive approach to exporting advanced AI chips to China.
For Anthropic, the 19-day shutdown carried real commercial cost. The terms it accepted to end it, the addition of strict classifiers, government stress-testing, and an ongoing information sharing arrangement, suggest the company’s most capable models will now operate under closer federal oversight going forward.
