Anthropic under the White House's wing: Mythos 5 is back in service.
The Trump administration has officially overturned a two-week ban on using Anthropic’s flagship cyber-models. Mythos 5 and Fable 5 are returning to the field, but not for everyone: access is restricted to an "inner circle" of 100 government agencies and loyal corporations. In a letter to Tom Brown, Anthropic’s Head of Compute, Commerce Secretary Howard Lutnick confirmed the White House has established "appropriate guardrails" to allow trusted partners to exploit the models for protecting critical infrastructure. Effectively, we are witnessing the transformation of a commercial startup into a restricted-access facility under Washington’s direct supervision.
This regulatory U-turn resolves an absurd conflict triggered by the June 12 ban, which had blocked even Anthropic’s own foreign employees from accessing the software. According to Semafor and Reuters, the new directives introduce exemptions for non-U.S. personnel within designated organizations. Interestingly, while Mythos 5 returns to the hands of the "lucky hundred," the fate of Fable 5 remains murky. This version was previously marketed as ultra-secure—until security researchers cracked it within days. On X (formerly Twitter), Anthropic has already promised to "rapidly restore access" while continuing close consultations with the government.
The Dictatorship of White-Lists
This case sets a dangerous precedent for "whitelisting," where cutting-edge SOTA models cease to be market commodities and turn into state privileges. A rigid stratification is emerging: a narrow layer of entities receives the regulatory right to powerful AI, while the rest of the business community is left with a diet of outdated or "neutered" iterations. Essentially, the administration is testing a model of state-licensed intelligence under the guise of national security.
Attempting to lock technology inside government offices is unlikely to prevent the security bypasses that hackers already demonstrated on Fable 5. However, it will inevitably slow down corporate defense speeds for those outside the "Lutnick list."
Key Takeaways of the New US Policy:
Access to flagship Anthropic models is limited to a list of 100 trusted organizations. Restrictions on foreign Anthropic specialists working on classified projects have been lifted. The U.S. government is effectively assuming the role of a high-performance AI distributor.
As Anthropic becomes embedded in the U.S. defense stack, the rest of the market is receiving a clear signal: innovation is now distributed by quota, not by checkbook.