BlockBeats News, July 1st, according to DynaBeat Monitoring, Anthropic announced that the U.S. government has lifted the export control on Claude Fable 5 and Claude Mythos 5. Previously, on June 12th, the U.S. government imposed export control on these two latest models, requiring Anthropic to restrict access by foreign nationals; due to the immediate effect of the order and the inability to reliably verify user nationality in real time, Anthropic temporarily suspended all user access to the two models.

Anthropic stated that Fable 5 will be reopened to global users starting July 1st, covering the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. For Pro, Max, Team, and some Enterprise plan users, Fable 5 will be subject to a weekly usage cap of up to 50% until July 7th, after which it will be provided through usage quotas. Anthropic also mentioned that access will be restored on AWS, Google Cloud, and Microsoft Foundry as soon as possible.

Regarding Mythos 5, Anthropic indicated that in accordance with the U.S. government's approval on June 26th, access has been restored to a group of U.S. institutions, and it will continue to coordinate with the government to expand to a broader set of U.S. and international partners in the Glasswing program. Anthropic stated that Fable 5 and Mythos 5 were released on June 9th, both based on the same underlying model, with Fable 5 having stronger security features for general-purpose use; Mythos 5 has less protection and is only available to a few trusted Project Glasswing partners for defensive cybersecurity scenarios.

Anthropic also mentioned that in the past two weeks, they have reviewed relevant reports with the U.S. government, Amazon, and other partners, and trained a new security classifier to intercept mentioned bypass behaviors in the reports. The new classifier can block relevant technologies in over 99% of cases, but it may also result in more benign requests being flagged during routine coding and debugging tasks. Anthropic stated that they will collaborate with Amazon, Microsoft, Google, and other Glasswing partners to develop an AI model "jailbreak" severity assessment framework and strengthen cooperation with the U.S. government in testing, information sharing, and research collaboration before model release.