Anthropic Relaunches Fable 5 After Lifting of U.S. Restrictions

Anthropic Relaunches Fable 5 After Lifting of U.S. Restrictions

After several weeks of negotiations with the Trump administration, Anthropic is finally able to bring Claude Fable 5 back online. In a post on X, Anthropic announced that it plans to start restoring access on Wednesday for users worldwide on the Claude platforms, and that the company will soon reactivate access on AWS, Google Cloud, and Microsoft Foundry, although no specific timeline was provided.

Anthropic stated:

"We have received notification that the Department of Commerce has lifted the export controls on Claude Fable 5 and Mythos 5. We will begin restoring access tomorrow and will share an update soon. We are grateful to our users for their patience, and to everyone who has worked with us to redeploy the models."

The company also published a blog post on Tuesday evening detailing the events that led to this situation, its new security measures, the new AI industry processes it is working on, as well as new ways to share information with the government, and pre-testing for upcoming models.

In early June, Anthropic had sidelined Fable 5 — its consumer-facing model built from the same underlying technology as Mythos 5, but with more protections — after an ultimatum from the Trump administration. Due to concerns about potential jailbreaks of the technology, the government imposed an export control directive on Anthropic, prohibiting any foreign national (including non-U.S. members of client companies and even many Anthropic employees) from using Mythos 5 or Fable 5, which were the two models Anthropic had highlighted the previous week.

To address the issue of jailbreaks, which had been flagged by Amazon researchers and significantly contributed to the implementation of the export control directive, Anthropic noted in its blog post that it had "trained an enhanced security classifier that targets and blocks" this behavior, adding: "Users will be informed if a request to Fable 5 is blocked, and the request will then be redirected to Opus 4.8. The new classifier means that the specific technique described in the Amazon report is blocked in over 99% of cases."

The Trump administration recently gave the green light for the return of Mythos 5, but only for a list of pre-approved organizations. Non-U.S. members of these organizations, as well as foreign employees of Anthropic, were able to regain access to the model. This decision came shortly after OpenAI unveiled GPT-5.6, which the Trump administration only allowed to be launched under similar rules: a phased deployment, initially reserved for a list of pre-approved organizations and government departments.

On Tuesday, Anthropic indicated that it "will continue to coordinate with the government to expand access to a broader set of national and international partners" for Mythos 5. The company also included an entire section in its blog post about its new plan for close collaboration with the Trump administration, highlighting its numerous attempts to regain the government's trust after months of public drama, legal battles, and presidential actions.

Anthropic stated that it plans to offer "government access and assessment in pre-launch," particularly for models relevant to national security capabilities, so that government partners can conduct independent assessments of the models' capabilities and test safeguards before broader deployment. (The government will also have access to Anthropic's technical staff during these pre-testing periods.)

Anthropic also plans to introduce "rapid information sharing" when "significant jailbreaks or patterns of abuse are identified."

It announced that it would work with the government and other leading AI labs to create a "shared, voluntary security and assessment standard" for advanced model providers. Finally, Anthropic stated that it would establish dedicated teams to work on shared government priorities, provide significant computing allocation to support government testing and research, and make its expertise in security and stress testing available to advance the state of the art in AI evaluation.

The initial export control directive from the Trump administration came at an inopportune time for Anthropic, as the company was preparing for an IPO and had been in disputes with the government for months regarding a risk designation for the supply chain.

The company emphasized in its blog post that there is "currently no consensus in the AI industry" on how to determine the severity of a jailbreak, a problem that "will become more acute in the coming months as more models with powerful cybersecurity (and other) capabilities are trained, evaluated, and released." It has therefore collaborated with Amazon, Microsoft, Google, and other companies participating in its Project Glasswing program to develop a widely accepted framework for assessing AI jailbreaks, proposing four categories:

• capability gain for the attacker
• extent of capability gain for the attacker
• ease of weaponization more generally
• discoverability (or how easily someone else can reproduce it)

Anthropic has also created a new team to "provide 24/7 monitoring of key jailbreak submission channels" and will soon launch a HackerOne program for researchers to submit potential jailbreaks they have reported for Fable 5.

Anthropic also included a warning in its blog post, writing that it is "probably impossible to make an AI model completely robust (i.e., impervious) to jailbreaks. We expect that some jailbreaks will be discovered for our models, and that they will vary in severity: there will be many minor jailbreaks, some narrowly harmful, and while no universal jailbreak for Fable 5 has been discovered at the time of writing, expert security researchers continue to test it."