Commerce Dept lifts Anthropic AI export ban after two-week standoff

The Commerce Department has lifted export restrictions on Anthropic's Mythos 5 AI model after two weeks of intensive negotiations over cybersecurity controls, allowing the company to restore access to more than 100 approved partners including federal agencies and private companies.

Commerce Secretary Howard Lutnick notified Anthropic on Friday that the company had made "significant progress" addressing government concerns about the security of its AI systems. The move partially resolves a dispute that began when the administration banned foreign nationals from using Anthropic's Fable and Mythos models over questions about whether the company's safety guardrails could prevent the technology from being used to launch cyberattacks.

Why it matters

The episode reveals the Trump administration's willingness to impose sudden restrictions on AI companies despite its stated goal of unleashing American dominance in the technology. With no clear framework for evaluating security concerns, leading AI developers now face uncertainty about whether their products will face similar government intervention—potentially slowing releases while Chinese competitors advance rapidly.

Limited access, broader questions

Anthropod can now redeploy Mythos 5 to "a small group of cyber defenders and infrastructure providers," the company said Friday. However, its general-purpose Fable 5 model remains unavailable for broader use. The company had voluntarily cut off all customer access after the initial restrictions to ensure compliance, while maintaining the situation represented "a misunderstanding."

According to POLITICO, which first reported the details, ongoing discussions between Anthropic and the administration aim to establish a standardized framework for evaluating suspected security bypasses in future cases. A person familiar with the talks said conversations were expected to continue through the weekend.

Industry pushback on approval process

The Anthropic case is not isolated. OpenAI announced Friday it is making only a limited release of three versions of its new GPT-5.6 model at the administration's request. In a blog post, OpenAI warned against making government access processes "the long-term default," arguing such requirements "keep the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."

The administration's approach has created confusion less than four weeks after President Trump signed an executive order rejecting mandatory federal AI controls in favor of voluntary government review. The rapid shift from voluntary to restrictive measures has raised concerns about whether the White House is unfairly targeting specific companies, particularly after previous clashes with Anthropic.

International implications

The export restrictions left U.S. allies in Europe and Canada suddenly unable to use Anthropic's models to identify and patch their own cybersecurity vulnerabilities. The disruption has renewed calls within the European Union to reduce dependence on American AI technology.

Meanwhile, reports of rapid advances in Chinese AI development have intensified industry worries that U.S. companies may lose their competitive edge while waiting for administration approvals.

Anthropod CEO Dario Amodei and President Trump sat across from each other during a G7 lunch with allied leaders and AI executives in Évian-les-Bains, France, last week, following the initial dispute.

POLITICO first reported the Commerce Department's reversal and the details of ongoing negotiations between Anthropic and the administration.