Anthropic said the US government has lifted the export controls that forced it to pull Claude Fable 5 and Mythos 5 from worldwide access, clearing the way for both models to return just under three weeks after they were suspended.
The suspension began June 12, when the government applied export controls after Amazon reported a jailbreak that could get Fable 5 to identify software vulnerabilities and produce exploit code. Because Anthropic could not reliably separate foreign users from domestic ones, the order forced it to disable both models worldwide, including for paying US customers, and set off talks between Washington and allied governments over how vetted partners might regain access.
Anthropic said it resolved the underlying issue by building an improved safety classifier that specifically targets the Amazon-reported bypass technique, blocking it in more than 99 percent of cases while still allowing through requests that lack enough detail to aid an attacker. Researchers at the Commerce Department's Center for AI Standards and Innovation tested both the prior and new safeguards and called them extraordinarily strong. The export controls were removed June 30, and Fable 5 becomes available globally across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork starting July 1. Mythos 5 access was restored earlier, on June 26, but only for a small set of trusted cyber-defense and infrastructure partners under Project Glasswing, Anthropic's own cybersecurity early-access program; it has not returned to general availability.
Anthropic framed its government commitments as an expansion of existing cooperation rather than a new concession. The company said it has worked with US government partners on pre-deployment testing and evaluation for nearly two years, and it was already one of the labs that had opted into the voluntary pre-release review the June 2 executive order set up for frontier models, ahead of the suspension. What is new is the scope: Anthropic said it would give designated government partners expanded early access to future models and safeguards, notify government counterparts quickly when significant jailbreaks or misuse patterns turn up, staff dedicated teams on shared government security priorities, and work toward a shared, voluntary security and evaluation standard across the industry. Separately, Anthropic, Amazon, Microsoft, Google, and other Glasswing partners are drafting a common framework for rating the severity of AI jailbreaks, based on the capability gain they produce, how broadly that capability applies, how easily it can be weaponized, and how discoverable it already is elsewhere.
The episode is the clearest resolution yet of the pre-release review regime the administration has been building around frontier models since June, in which the government can gate access to a model on national-security grounds and labs work their way back through cooperation rather than court challenges. Anthropic, unlike Meta, had already bought into that framework before the suspension; what this deal adds is a validated technical fix, an expanded version of the access arrangement, and a new industry framework for grading jailbreaks that did not exist three weeks ago.
Sources: Anthropic, CNBC, Cybersecurity Dive
–
By the Control Plane Editorial Team