Export controls lifted; Anthropic restores Claude Fable 5 and Mythos 5
Export controls on Claude Fable 5 and Claude Mythos 5 were removed on June 30, leading to the global return of Fable 5 on July 1 across Anthropic’s platforms and a restricted restoration of Mythos 5 to a set of U.S. organizations. The freeze began on June 12 when the government applied the controls, an order that took effect immediately and suspended access for all users.
On June 12 the government applied export controls to Anthropic’s Claude Fable 5 and Claude Mythos 5, and the order took effect immediately, suspending access for all users. The immediate suspension removed access to both models worldwide. The freeze interrupted normal availability of Fable 5 and Mythos 5 across Anthropic’s platforms.
Following the freeze, the government approved restoring Mythos 5 to a set of U.S. organizations on June 26, enabling restricted access for those organizations. Export controls were removed on June 30. After the removal, Claude Fable 5 returned to global availability on July 1 across Anthropic’s platforms. Mythos 5 remained restored only to certain U.S. organizations rather than resuming global availability.
These developments sequence the initial June 12 order, the June 26 partial restoration for Mythos 5, the June 30 removal of controls, and the July 1 global return of Fable 5. Access status changed from full suspension to selective restoration and then broad resumption.
Amazon researchers reported a jailbreak that could bypass safeguards on Claude Fable 5, and a cybersecurity finding followed that report. The reported jailbreak demonstrated methods that could circumvent the model’s built-in protections. Such jailbreaks can prompt the model to identify software vulnerabilities and, in one case, produce exploit code. The incident was recorded as a cybersecurity finding.
Amazon researchers’ report directly led to the cybersecurity finding, establishing the researchers’ role in identifying the issue. Anthropic is drafting a framework with Amazon, Microsoft and Google for scoring how dangerous a jailbreak is. Anthropic has said it is deepening collaboration with the U.S. government, including giving designated agencies early access to frontier models and safeguards. These factual items were recorded alongside the models’ security review in the documented timeline.
Anthropic said it is deepening collaboration with the U.S. government and with major technology companies Amazon, Microsoft and Google, and it described specific measures in support of that effort. As part of those measures, Anthropic is giving designated agencies early access to frontier models and to associated security safeguards. Anthropic is also drafting a framework with Amazon, Microsoft and Google for scoring how dangerous a jailbreak is.
These arrangements tie into a June 2 executive order on AI security and suggest a tighter link between frontier AI releases and government review. The drafting of a shared scoring framework is described as a means to evaluate the danger posed by jailbreaks across frontier models. Anthropic’s stated early-access measures and the joint framework were recorded alongside other steps the company cited as deepening collaboration with government agencies.
Anthropic’s AI models were suspended after government-imposed export controls and have since been restored, with Claude Fable 5 returning to global availability and Claude Mythos 5 restored to selected U.S. organizations.
A cybersecurity finding followed after Amazon researchers reported a jailbreak that could bypass safeguards and potentially prompt models to identify software vulnerabilities or produce exploit code, and Anthropic says it is collaborating with the U.S. government and major tech companies to provide early access and develop a shared framework to score the danger of jailbreaks.
