Anthropic has announced that it will make its latest model, Claude Fable 5, available again to users worldwide starting July 1 (Japan time), after the US government lifted the export controls that had applied to it[1]. Access to the model had been suspended for all users following export controls that took effect on June 12, but it can now be used again. The company also shared details on strengthened safeguards and closer collaboration with the US government.
From the June 12 Suspension to Restoration
Anthropic released Claude Fable 5 and Claude Mythos 5 on June 9[1]. On June 12, however, the US government applied export controls to the two models, requiring access to be restricted for foreign nationals both inside and outside the United States. Because the order took effect immediately and there was no reliable way to verify nationality in real time, Anthropic temporarily suspended access to both models for all users[1].
The export controls on both models were then lifted as of June 30[1][2]. Fable 5 is available from July 1 to users worldwide on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. For the Pro, Max, Team, and select Enterprise plans, it is included for up to 50 percent of weekly usage limits through July 7, after which it is offered via usage credits[1]. Availability on AWS, Google Cloud, and Microsoft Foundry will be restored as quickly as possible.
The more capable Mythos 5, meanwhile, was restored for a set of US organizations following US government approval on June 26[1].
Triggered by a Report of a Safeguard Bypass
The export controls were prompted by a reported method for bypassing Fable 5's safeguards[1]. Researchers at Amazon discovered a way to prompt Fable 5 into identifying software vulnerabilities, and in one case had it generate code demonstrating how a vulnerability could be exploited.
According to Anthropic's own testing, however, less capable models such as Claude Opus 4.8, GPT-5.5, and Kimi K2.7 could identify the same vulnerabilities[1]. Every model tested could also reproduce the same exploit demonstration, so the report did not expose any capability unique to Fable 5, the company said.
A New Classifier That Blocks Over 99 Percent
Working with the government, Anthropic trained an improved safety classifier that specifically targets and blocks the reported behavior[1]. As a result, the reported technique is blocked in over 99 percent of cases. Blocked requests are flagged to the user and routed instead to Claude Opus 4.8. CAISI, the Center for AI Standards and Innovation under the US Department of Commerce, also evaluated both the old and new safeguards and judged them to be extraordinarily strong.
At the same time, the new classifier more frequently flags harmless requests during routine coding and debugging. Anthropic says it will keep improving its ability to distinguish genuine misuse from legitimate requests.
Toward a Common Industry Framework
Anthropic noted that the industry still lacks a common standard for objectively measuring the severity of an AI "jailbreak" (a technique that bypasses a model's safeguards), and said it has begun drafting an assessment framework together with Amazon, Microsoft, Google, and others[1]. The current proposal scores a jailbreak on four criteria: capability gain, the breadth of that gain, ease of weaponization, and discoverability. The company is also launching a HackerOne program where researchers can report jailbreaks they find.
Summary
Claude Fable 5, which had been suspended for all users under the June 12 export controls, is being made available again worldwide from July 1 following the lifting of the controls on June 30. With stronger safety classifiers, a push for a common industry assessment framework, and deeper collaboration with the US government, Anthropic's effort to balance access to high-performance models with safety looks set to continue.
