After weeks of negotiations with the Trump administration, Anthropic will finally be able to bring Claude Fable 5 back online. In a post on X, Anthropic said it plans to begin restoring access tomorrow to users around the world on the Claude platform, and that the company will re-enable access to AWS, Google Cloud, and Microsoft Foundry soon, but without a set timeline.
We have received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5.
We will start restoring access tomorrow, and will share an update soon.
Thanks to our users for their patience, and everyone who worked with us on the re-release of the models.
The company also published a blog post on Tuesday evening detailing the lead-up to the events in question, its updated defenses, the new AI industry processes it is working on, and the new ways it plans to share information with the government, as well as pre-testing the release of future models.
At the beginning of June, Anthropic put aside the Fable 5 – its consumer-oriented model built on the same basic technology as Anthropic’s Mythos 5, but with additional protections – after the end of Friday evening from the Trump administration. Because of concerns about possible jailbreaking of the technology, the government hit Anthropic with an export control order, not allowing any foreign person (including non-US client members and many Anthropic employees) to use the Mythos 5 or the Fable 5, that is, both models that Anthropic spent the last week discussing.
Because of concerns about potential jailbreaks, the government hit Anthropic with an export control order, affecting both models that Anthropic spent last week talking about.
To deal with the jailbreak in question, which was flagged by Amazon researchers and in particular who were responsible for setting up the export control order, Anthropic said in a blog post that “it has trained an advanced security unit that directs and prevents” that behavior, adding that, “Users will be notified if the Fable 5 application is blocked, and the application will be sent to the new Opus in class 4. in more than 99% of cases.
The Trump administration recently greenlit the return of Mythos 5, but only to a pre-approved list of organizations. Non-US members of those organizations, as well as Anthropic employees in other countries, are allowed to regain access to the model. This decision came shortly after OpenAI launched GPT-5.6, which the Trump administration only allowed to launch under the same rules: a surprise release, initially a pre-approved list of organizations and government departments.
On Tuesday, Anthropic wrote that the company “will continue to engage with the government to expand access to a broader set of domestic and international partners” in Mythos 5. The company also included an entire section in its blog post about its new plan to work closely with the Trump administration – highlighting its many efforts to return to government programs after months of public drama, lawsuits, and presidential action.
Anthropic’s blog post includes an entire section of its new action plan around the Trump administration.
The company wrote that it plans to provide “pre-release government access and testing,” especially for models that are relevant to national security forces, so that government partners can conduct independent tests on the model’s capabilities and test measures before a wider release. (The government will also have access to Anthropic’s technical staff during those pre-release testing periods.) Anthropic also said it plans to introduce “rapid information sharing” when “serious prison violations or patterns of abuse are identified.”
It said it would work with governments and other leading AI labs to create a “shared, voluntary and testable security standard for boundary model providers.” Ultimately, Anthropic said it will “stand up dedicated Anthropic teams to work on shared government priorities, provide critical computing resources to support government testing and research, and make our security and red teaming expertise available to help advance the state of the art in AI testing.”
Anthropic said it will work with governments and other leading AI labs to create a “voluntary” shared standard for security and testing.
The first export control order by the Trump administration comes at an inopportune time for Anthropic, as it prepares for an IPO and has been at loggerheads with the government for months over supply chain risk designations.
The company highlighted in a blog post that “currently there is no consensus in the AI industry” about deciding on the difficulty of the prison, and that it is an issue that “will become more difficult in the coming months, as more models with strong cybersecurity skills (and others) are trained, tested, and released.” So it said it is partnering with Amazon, Microsoft, Google, and other companies that are part of its Project Glasswing program to write a widely agreed upon framework for assessing risks. of AI prisons, with four categories proposed: attacker’s leverage, attacker’s scope of advantage, weapon’s general usability, and ease of access to another person (or ease of recapture). Anthropic said it has created a new team that will “provide 24/7 monitoring of key jailbreak transmission channels” and will soon launch a HackerOne program for researchers to submit potential jailbreaks to Fable 5.
Anthropic said it has created a new team that will “provide 24/7 monitoring of the prison’s critical transmission channels.”
Anthropic also included a disclaimer in its blog post, writing that “it is almost impossible to make any AI model fully robust (ie, invincible) to jailbreak. We expect that some jailbreaks will be found in our models, and that they will vary in severity: there will be many minor jailbreaks, some devastating, and although there are no security experts worldwide. researchers are still compiling them.”
