Anthropic has announced that it is redeploying Claude Fable 5 globally starting July 1, after the US government lifted export controls that had temporarily restricted access to the company’s newest AI models. The move comes just over two weeks after Washington ordered Anthropic to limit availability of both Fable 5 and Mythos 5 to foreign nationals.
What happened? #
On June 12, the US government imposed export controls on Fable 5 and Mythos 5. Because the order took effect immediately and Anthropic had no reliable way to verify user nationality in real time, the company suspended access to both models for all users.
As of June 30, those controls have been lifted. Fable 5 will become available again on July 1 across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Anthropic also said it will restore cloud access via AWS, Google Cloud, and Microsoft Foundry as quickly as possible.
Who gets access? #
For Pro, Max, Team, and select Enterprise subscribers, Fable 5 will be included for up to 50% of weekly usage limits through July 7. After that, access will move to a usage-credits model. Mythos 5 — the less-restricted variant intended for defensive cybersecurity work — is being restored for a limited set of US organisations approved by the government on June 26. Anthropic is working to expand Mythos 5 access to more domestic and international partners in its Glasswing programme.
New safeguards added #
The export controls followed an Amazon research report that described a way to bypass Fable 5’s safety guardrails. The technique prompted the model to identify software vulnerabilities, and in one case produced code showing how a vulnerability could be exploited.
Anthropic said the reported method did not expose any unique offensive capabilities of Mythos 5. In its own tests, several smaller models — including Claude Opus 4.8, GPT-5.5, and Kimi K2.7 — could identify the same vulnerabilities, and every model tested could produce the same exploit demonstration.
Even so, Anthropic trained an improved safety classifier that now blocks the reported behaviour in more than 99% of cases. If a request is blocked, users will be notified and redirected to Claude Opus 4.8 instead. The company noted the tighter classifier may also flag more benign coding and debugging requests, and it plans to continue refining it.
A proposed industry framework for jailbreaks #
The episode prompted Anthropic to call for a shared industry framework to assess AI jailbreak severity. Along with Amazon, Microsoft, Google, and other Glasswing partners, it proposed scoring jailbreaks on four criteria:
- Capability gain — how far beyond existing tools the jailbreak takes an attacker
- Breadth of capability gain — how many distinct offensive tasks the technique enables
- Ease of weaponization — how much effort is needed to turn the technique into an attack
- Discoverability — how easily the technique can be found and copied
The company said it will also launch a HackerOne programme so security researchers can submit potential cyber jailbreaks in Fable 5 for review.
Deeper US government collaboration #
Anthropic outlined plans to strengthen coordination with the US government, including:
- Expanded pre-release government access and independent evaluation of frontier models
- Rapid information sharing on jailbreaks, misuse patterns, and new safeguards
- Dedicated Anthropic teams, compute allocations, and red-teaming expertise for joint research
- Working toward a common, voluntary security and evaluation standard across frontier model providers
Source: Anthropic