The Trump administration has partially reversed its ban on Anthropic's Mythos 5 model, granting access to more than 100 US government agencies and companies — including their non-American employees — two weeks after forcing the model off the market, according to TechCrunch.
Commerce Secretary Howard Lutnick informed Anthropic on Friday that "appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model," according to a letter seen by Semafor. Reuters confirmed the partial reversal, noting that the administration did not extend the same approval to Fable 5 — the more restricted variant of Mythos 5 that was pulled days before the ban after security researchers allegedly bypassed its guardrails.
Anthropic acknowledged the progress in a post on X, saying it was "restoring access for these organizations quickly" while continuing to work with the government to expand access further and bring Fable 5 back to general availability. The company's own non-American employees, who were caught in the original ban, are now included in the approved access list.
Analysis
The partial reversal looks like a compromise, but it is actually a hardening of the new regime. The government has now demonstrated it can revoke access to a frontier model on national security grounds, force a two-week blackout, and then selectively restore access to a vetted list of "trusted partners" — all without legislation, without judicial review, and without public criteria for who makes the list. The 100 organisations on Lutnick's list are now inside a de facto security-cleared AI access tier. That creates two classes of companies: those the government trusts with frontier models and those it doesn't. In 12 months, being on that list will be a competitive moat worth billions. Being off it — whether by government decision or by choosing not to seek approval — will be a liability. The export controls saga with Anthropic's non-American users was the opening act. The main event is the quiet construction of a government-gated market for frontier AI access, and every company in the industry is watching to see whether OpenAI's staggered GPT-5.6 rollout follows the same template.
