The U.S. government is taking the next step in its oversight of the most powerful AI models. Following the dispute with Anthropic over the export of Fable 5 and Mythos 5, more than 100 selected U.S. organizations are now permitted to use Mythos 5 again. This marks a shift in policy from a blanket ban to controlled access.
As Techzine previously reported, the Trump administration imposed export restrictions on Anthropic on June 13. The company was required to restrict access to its most advanced models, Fable 5 and Mythos 5, for foreign users. Because Anthropic did not have a robust system in place at the time to distinguish between U.S. and non-U.S. users, it took both models offline worldwide.
Just over a week later, President Donald Trump stated in an interview with Axios that he no longer viewed Anthropic as a threat to national security. He also praised the company’s swift response to the U.S. government’s concerns. At the same time, Anthropic announced that it would introduce identity verification for certain Claude users to ensure better compliance with future export regulations.
Mythos 5 available again
Anthropic announced on Friday that the U.S. government has granted permission to make Mythos 5 available again to a limited group of U.S. organizations. According to the company, these are entities that manage or protect critical infrastructure, Reuters reports.
According to sources, more than 100 companies and institutions will gain access to the model, including several Fortune 500 companies. Many of these organizations are part of Project Glasswing, a program through which Anthropic makes its most advanced AI models available on a controlled basis to technology companies, financial institutions, and other strategic partners.
From export ban to authorization policy
This partial release demonstrates that Washington is further refining its approach. Whereas the initial measures were primarily aimed at preventing the models from leaving the United States, a system now appears to be emerging in which the government determines which organizations gain access to the most powerful AI systems.
This is also evident from a letter from U.S. Secretary of Commerce Howard Lutnick. The letter states that approved companies no longer need an export license when employees who are not U.S. citizens use Mythos 5. The exception also applies to foreign employees of Anthropic itself. However, organizations not on the approved list remain subject to existing export restrictions.
This effectively creates an authorization model in which the government not only determines which AI models are released but also who is permitted to use them.
Anthropic is not the only company facing this new approach. On Friday, OpenAI announced that it would postpone the widespread rollout of GPT-5.6 at the request of the U.S. government. For the time being, that model will also be made available only to a limited group of pre-vetted partners.
According to the government, these measures are necessary to prevent advanced AI models from being used for offensive cyber operations or other applications that could harm national security. In particular, the government highlights the risk of such technology falling into the hands of state actors in countries such as China and Russia.
Criticism of the government’s selection process
Meanwhile, the new approach has drawn criticism. Legal experts and advocacy groups argue that it is unclear what criteria are used to select organizations. As a result, they contend, the government gains significant influence over the availability of the most powerful AI models without a transparent evaluation process to support it.
OpenAI CEO Sam Altman also voiced criticism of this development. He said he understands the need for extensive safety testing before new models are made widely available, but believes the government should not be given the role of determining which organizations gain access.
Fable 5 may soon get the green light
Moreover, the easing of restrictions does not appear to be over yet. Reuters reported on Saturday, citing a story from Axios, that the Trump administration is close to reaching an agreement to make Fable 5 available again as well. Reuters has not yet been able to independently confirm that report. Neither Anthropic nor the White House would comment to Reuters at this time.
If Fable 5 is also released, the export crisis from mid-June appears to have largely been transformed into a new U.S. access policy. The focus is no longer on a complete blockade, but on a model in which the government determines which organizations gain access to the most advanced AI systems and under what conditions that occurs.