Anthropic is preparing to bring Claude Fable 5 back online after weeks of negotiations with the Trump administration. The return follows notice that export controls on Claude Fable 5 and Mythos 5 have been lifted, allowing Anthropic to start restoring access on Claude platforms for users globally.
Access is coming back in stages
Anthropic said in a post on X that it plans to begin restoring access Wednesday to users globally on Claude platforms. The company also said it would re-enable access on AWS, Google Cloud, and Microsoft Foundry soon, though it did not give a fixed timeline for those platforms.
The company wrote:
We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5.
It also said:
We’ll begin restoring access tomorrow, and will share an update soon.
The move brings back a consumer-facing Mythos-class model that Anthropic had sidelined in early June. Fable 5 is built from the same underlying technology as Anthropic’s Mythos 5, but the source describes it as having more safeguards.
Why Fable 5 was sidelined
In early June, Anthropic pulled Fable 5 back after a Friday evening ultimatum from the Trump administration. The issue centered on concerns over potential jailbreaks of the technology.
The government had issued an export control directive that blocked any foreign national from using Mythos 5 or Fable 5. That restriction included non-US members of enterprise client companies and many of Anthropic’s own employees.
The timing was difficult for Anthropic. The source says the directive arrived as the company was preparing for an IPO and after months of conflict with the government over a supply chain risk designation.
The restriction also applied to both models Anthropic had spent the previous week promoting. Mythos 5 has already been allowed to return, but only to a preapproved list of organizations. Non-US members of those organizations and Anthropic’s own foreign national employees were allowed to regain access to that model.
The new safeguards around jailbreaks
Anthropic’s blog post laid out how the company says it addressed the jailbreak concern. The specific jailbreak had been flagged by Amazon researchers and was described as largely responsible for setting the export control directive in motion.
Anthropic said it had “trained an improved safety classifier that targets and blocks” the behavior. If a request to Fable 5 is blocked, users will be notified and the request will instead be sent to Opus 4.8.
The company also said:
The new classifier means that the specific technique described in the Amazon report is blocked in over 99% of cases.
That does not mean Anthropic is claiming the model is impossible to jailbreak. In fact, its blog post included a clear disclaimer that full robustness is probably unattainable for any AI model.
Anthropic wrote that “it is probably impossible to make any AI model fully robust (that is, impervious) to jailbreaks.” It also said it expects some jailbreaks to be found, with severity ranging from minor to harmful, while adding that no universal jailbreaks for Fable 5 had been discovered at the time of writing.
A closer relationship with government testing
The return of Claude Fable 5 is not only about one model coming back online. Anthropic also described a broader plan for how it will work with the Trump administration on AI evaluation, prerelease testing, and information sharing.
The company said it plans to offer “pre‑release government access and evaluation,” especially for models relevant to national security capabilities. Government partners would be able to run independent evaluations of model capabilities and test guardrails before a wider release.
During those prerelease testing periods, the government would also have access to Anthropic’s technical staff. Anthropic said it plans to introduce “rapid information sharing” when “significant jailbreaks or misuse patterns are identified.”
The company also said it would work with the government and other leading AI labs to create a “shared, voluntary security and evaluation standard for frontier model providers.” In addition, Anthropic said it would set up dedicated teams for shared government priorities, provide a significant compute allocation for government testing and research, and make its safety and red-teaming expertise available.
What this says about frontier model release
Anthropic’s update points to a larger problem for AI companies: there is no settled industry standard for judging how severe a jailbreak is. The company wrote that there is “currently no consensus in the AI industry” for making that determination.
Anthropic said the issue will become more urgent as models with powerful cybersecurity and other capabilities are trained, assessed, and released. To address that, it said it partnered with Amazon, Microsoft, Google, and other enterprises in its Project Glasswing program to draft a framework for evaluating AI jailbreaks.
The proposed framework has four categories:
- Capability gain for the attacker
- Breadth of capability gain for the attacker
- Ease of weaponization more broadly
- Discoverability, or how easy it may be for someone else to repeat it
Anthropic also said it created a new team to “provide 24/7 monitoring of key jailbreak submission channels.” It plans to debut a HackerOne program for researchers to submit potential jailbreaks they have flagged for Fable 5.
For users, the practical result is that Claude Fable 5 is returning first through Claude platforms, with cloud access to follow later. For the AI industry, the episode shows how model releases, jailbreak research, export controls, and government evaluation are becoming increasingly connected.