The US government has granted Anthropic permission to release Claude Mythos 5, its most advanced cybersecurity model, to a select group of over 100 US organizations. Commerce Secretary Howard Lutnick wrote in a letter dated Friday that he had “determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,” as reported by Reuters. The partial reinstatement comes roughly two weeks after the White House ordered an export block on both Mythos and its consumer-facing sibling Fable 5, citing national security concerns.
What is genuinely new here is not that the government blinked. It is the mechanism. The Trump administration did not grant a blanket release. It created a tiered access system for a single frontier model, with explicit government sign-off on who gets to use it. Lutnick’s letter, obtained by WIRED, specifies that “organizations approved to use Mythos may now allow their foreign national employees to access the model, and Anthropic may do the same for its own foreign national employees.” That is a carve-out from the original June 12 directive, which required Anthropic to limit foreign nationals from accessing the models entirely, including employees living and working in the US.
The government stopped short of permitting a broader rollout. Fable 5, the consumer-facing version with additional safeguards, remains blocked. “The letter does not include permission for Anthropic to release Fable,” CNN reported. Anthropic spokesperson Eduardo Maia Silva confirmed the company is “working to provision the approved set of providers and restore their access to Mythos 5 as quickly as possible.” Conversations between Anthropic and the government are expected to continue into the weekend, with an eye to restoring access to Fable as well, according to a source familiar with the discussions.
This is not a clean resolution. It is a messy compromise that reveals how unprepared the US regulatory apparatus is for the era of frontier AI models.
The backstory matters. The Trump administration’s concern about Mythos crystallized after Anthropic granted access to a South Korean telecommunications firm the US believed had ties to China, WIRED reported. Separately, Amazon and the National Security Agency raised concerns that Fable 5 could be jailbroken. The confluence of events convinced officials they needed to act. On June 12, the White House sent an export control directive to Anthropic requiring the company to suspend all use by foreign nationals. Anthropic responded by disabling customer access to both models entirely.
The result was a mess. Anthropic’s own foreign national employees could not access the models they built. The company’s investors worked through the weekend to determine what the directive meant for Anthropic’s corporate future, WIRED reported. The episode underscored the lack of a consistent regulatory framework around AI, even as the technology advances rapidly and the US tries to stay ahead of global competitors like China.
The partial reinstatement of Mythos is a step forward for Anthropic, but it raises a troubling precedent. The government now effectively has a veto over which organizations can access which frontier models, and on what terms. Dean Ball, head of the strategic futures team at OpenAI and a former White House AI adviser, said in a June 16 blog post that Anthropic’s spat with the administration has shown frontier AI model developers they “need an explicit green light from the government now.”
That is the new reality. Not a formal regulatory framework debated in Congress. Not a clear set of rules published in the Federal Register. But a phone call, a letter from the Commerce Secretary, and a carve-out for “trusted partners” that the government gets to define. OpenAI announced Friday it was delaying the release of its upcoming GPT 5.6 models in response to a similar request from the Trump administration.
The Mythos precedent is bad for builders. It creates uncertainty. Every frontier lab now knows that the government can, at any moment, order a model taken offline. The criteria for that decision are opaque. The timeline for resolution is unpredictable. And the cost of compliance is borne entirely by the company. Anthropic has already sued the Trump administration over a separate “supply chain risk” designation it received after trying to draw red lines around how military contractors could use its models. The company has notched at least one early win in that case, but the legal fight is ongoing.
The government’s relationship with Anthropic has been particularly fraught. Earlier this year, the administration labeled the company a “supply chain risk,” effectively blacklisting it, over a disagreement about military uses for Anthropic’s products. Anthropic sued. Now the same government is the gatekeeper for which organizations can use Mythos.
What this means for the industry is a period of ad hoc, case-by-case regulation. The Trump administration has taken a largely hands-off approach to regulating artificial intelligence, seeking to keep US companies ahead of Chinese competitors. But the Mythos episode shows that hands-off does not mean no hands. When a model triggers enough concern, the government will act. And it will act through export controls, not through a public rulemaking process.
The outstanding question is whether this incident will help inform a lasting policy framework. Both parties are hopeful it will, according to the person familiar with the matter. But hope is not a plan. The US needs a clear, predictable system for evaluating and regulating frontier model releases. Without one, every new model launch will be a negotiation with the government, and every company will face the same uncertainty Anthropic faced this month.
For now, Mythos is back online for a small group of cyber defenders and infrastructure providers. Fable remains blocked. The government’s trust is conditional, and the conditions are not public. That is the new normal for frontier AI.