Anthropic restored global access to Claude Fable 5 and Mythos 5 on July 1, 2026, after the US government lifted export controls on June 30 — controls that had been in force since June 12. Together with the announcement, Anthropic published a detailed account of the models' safety mechanisms and a proposed industry framework for scoring jailbreak severity, developed with Amazon, Microsoft, Google, and Project Glasswing partners.
Key takeaways
- Claude Fable 5 available globally from July 1, 2026 on claude.ai, Claude Code, and Claude Cowork
- Export controls were in effect for 19 days (June 12–30), blocking access for both domestic and international users
- A new safety classifier blocks the described jailbreak technique in over 99% of cases
- Anthropic, Amazon, Microsoft, and Google propose a shared four-criteria framework for assessing jailbreak severity
- Anthropic commits to pre-release government testing for frontier models relevant to national security
Why the models were suspended
Export controls were imposed on June 12 after Amazon researchers discovered a technique that bypassed Fable 5's safety classifiers — the model identified several software vulnerabilities and, in one case, generated a demonstration of how a vulnerability could be exploited.
Anthropic then ran comparative tests across more than a dozen models. The results were clear: the same vulnerabilities were identified by Claude Opus 4.8, GPT-5.5, and Kimi K2.7 — and the exploitation demo for one of them could be generated by every model in the tested set, including Claude Haiku 4.5, Sonnet 4.6, Opus 4.6, Opus 4.7, and Opus 4.8, as well as GPT-5.4. Fable 5 offered no unique offensive capabilities unavailable in weaker models.
In response, Anthropic trained an improved safety classifier that blocks the described technique in over 99% of cases. Researchers from the Department of Commerce's Center for AI Standards and Innovation (CAISI) tested both the previous and updated safeguards and confirmed their effectiveness.
Proposed jailbreak severity framework
Both the suspension and reinstatement of the models exposed a gap the whole industry shares: no agreed-upon standard for objectively assessing jailbreak severity. Anthropic — together with Amazon, Microsoft, Google, and other Glasswing partners — proposes a four-criteria scoring system:
- capability gain
- breadth of capability gain
- ease of weaponization
- discoverability
The analogy is the Common Vulnerability Scoring System (CVSS) used to rate software vulnerabilities — a standard the AI industry has lacked. Alongside the framework, Anthropic is launching a new HackerOne program where security researchers can submit discovered jailbreaks in Fable 5 for review.
New commitments to the US government
Beyond the jailbreak framework, Anthropic announced four concrete commitments to the US government:
- pre-release government access for models materially advancing the capability frontier in national security areas
- rapid disclosure of significant jailbreaks
- dedicated teams and compute resources for joint government research
- participation in developing a voluntary industry security and evaluation standard for frontier model providers
Access after reinstatement
From July 1, Fable 5 is available globally on Pro, Max, Team, and select Enterprise plans — covering up to 50% of weekly usage limits through July 7, after which access requires purchasing usage credits. Access via AWS, Microsoft Foundry, and Google Cloud will be restored as quickly as possible. Mythos 5 access remains restricted to US organizations approved by the government under the Glasswing program. Claude Code users have access through claude.ai.
Why this matters
The suspension and reinstatement of Fable 5 exposed a gap affecting the entire industry — the absence of a shared vocabulary for assessing the severity of an AI jailbreak in a national security context. The proposed four-criteria framework is an attempt by private actors to fill that gap before legislation imposes its own, potentially less precise standards.
For organizations using AI models with sensitive data, Anthropic's public confirmation is also significant: the technique that triggered the suspension gave Fable 5 no unique capabilities unavailable in weaker models. That is important context for organizations reassessing whether returning to Fable 5 is justified.
Deeper US government involvement in pre-release model testing sets a precedent that other major labs — OpenAI, Google — will need to factor into their own plans for releasing frontier models.
What's next
- Anthropic plans to publish the detailed jailbreak framework after gathering feedback from additional industry partners — including firms outside Glasswing invited to join the effort.
- Fable 5 access via AWS, Google Cloud, and Microsoft Foundry is to be restored as soon as possible — no specific date given.
- The new HackerOne jailbreak submission program for Fable 5 is live at hackerone.com/anthropic-cyber-jailbreak.
