The Trump administration is walking back its ban on Anthropic’s most powerful cybersecurity model. Two weeks after a ban forced Anthropic to pull Mythos 5 and Fable 5 from the market, the government is now letting the company deploy Mythos 5 to more than 100 specific U.S. agencies and companies, according to TechCrunch AI, which cites reporting from both Semafor and Reuters. That includes a notable reversal: non-American employees at those organizations, and Anthropic’s own non-American staff, can now access the model after being locked out entirely under the original order.
“I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,” Lutnick wrote, in a missive seen by Semafor and reported by TechCrunch AI.
What actually changed
The shift is narrow but meaningful. Here’s what the new directive does:
- Restores Mythos 5 access for a defined list of 100-plus U.S. government agencies and companies.
- Specifically names organizations that “operate and defend critical infrastructure.”
- Lifts the blanket restriction on non-American employees at those orgs, including Anthropic’s own.
And here’s what it pointedly does not do: the administration didn’t address Fable 5 at all. Fable 5 is a version of Mythos 5 that Anthropic released a couple of days before the ban, billed as having stronger protections. Both models got pulled after security researchers reportedly bypassed those guardrails with little effort.
Why this matters
This is a rare case of the government un-banning a frontier AI model rather than tightening the screws. The original ban was a blunt instrument. It treated Mythos 5 as too dangerous to ship, full stop, after its safety guardrails failed under testing. The new approach is more surgical: keep the model out of general circulation, but hand it to a vetted set of critical-infrastructure defenders who arguably need offensive-grade cyber tools the most.
That trade-off tells you how Washington is starting to think about powerful AI. Instead of a binary release-or-block decision, regulators are experimenting with tiered access tied to who the user is and what safeguards are in place. For a model built to find and exploit security weaknesses, that distinction between “trusted defender” and “general public” carries real weight.
What stands out to me is the non-American employee piece. The original order barred non-Americans from touching the models, a restriction that hit Anthropic’s own workforce. Reversing it signals the government’s concern was less about nationality across the board and more about getting controlled deployment terms it could live with.
What comes next
Anthropic isn’t treating this as the finish line. The company acknowledged the progress publicly on X Friday: “Since June 12, we’ve been working closely with the US government to restore access to Claude Mythos 5 and Fable 5. Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure. We’re restoring access for these organizations quickly, and we’re continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.”
Read that carefully and the roadmap is clear. Anthropic wants two things from here: a wider list of approved organizations for Mythos 5, and a path to put Fable 5 back into general release. Neither is guaranteed. The guardrail failures that triggered the ban haven’t been publicly resolved, and Fable 5’s silence in the latest directive suggests the government still isn’t comfortable with broad availability.
For anyone building on or watching frontier models, this is a preview of how high-capability AI may get governed going forward: not a clean yes or no, but negotiated, conditional, and reviewed case by case. Anthropic did not immediately respond to a request for comment. More details are available at the original TechCrunch AI report.