Trump Administration Lifts AI Export Ban on Anthropic Model

The Trump administration has walked back its characterization of Anthropic as a national security threat, just days after imposing an export ban on the AI company's latest models. President Trump told Axios on Friday that Anthropic is no longer considered a threat, adding "not now, but a week ago, maybe."

The abrupt reversal follows a turbulent week that began when the government classified Anthropic's Fable 5 model as a national security risk after being notified of a jailbreak vulnerability. The administration gave the company just 90 minutes to pull both Fable 5 and its more advanced predecessor Mythos from all customers, according to a source close to the company. The export control even prevented some of Anthropic's own employees from accessing the models.

Anthropic, valued at nearly a trillion dollars and preparing for a public offering, maintains the vulnerability is minor and exists in competitors' models as well. The company received no specific details about the national security concerns that triggered the ban, according to its June 12 statement.

Why it matters

The episode reveals a fundamental gap in how the United States regulates frontier AI systems. Without transparent processes for evaluating AI risks, companies face unpredictable enforcement actions that could push development overseas while doing little to address genuine security concerns. As China and other nations advance their AI capabilities, an ad hoc regulatory approach may undermine rather than protect American technological leadership.

Broader regulatory tensions

The conflict extends beyond a single model. The Department of Defense had already blacklisted Anthropic as a "supply chain risk" after the company disagreed with Pentagon requests to modify its AI systems' guardrails for military applications. The company had also drawn scrutiny for Mythos, which it described as exceptionally capable at identifying security vulnerabilities—prompting Anthropic to limit the full version to select partners.

Amazon reportedly first flagged the jailbreak vulnerability to the government, according to a source familiar with the situation. Several prominent AI safety researchers, including former Facebook chief security officer Alex Stamos, reviewed the government's findings and concluded the capabilities did not justify such an extreme response. Dozens of cybersecurity researchers and AI executives signed an open letter Monday criticizing the lack of transparency and warning that restricting defensive tools while adversaries advance poses its own dangers.

Trump adviser David Sacks defended the administration's initial action, questioning how anyone could dismiss a jailbreak enabling "operability of a cyber weapon" as not serious.

The regulatory vacuum

The Trump administration has favored a light regulatory touch for AI, rolling back Biden-era mandatory safety reporting requirements in favor of voluntary frameworks. An executive order issued this month asks AI companies to voluntarily share advanced models with the government for cybersecurity vetting before public release—though Trump delayed the order over concerns it might hinder innovation.

Jessica Tillipman, associate dean for government procurement law at George Washington University, wrote that the problem is not government discretion itself but "the absence of any meaningful process." She warned that governing through opaque, personalized actions will ultimately trigger the heavy regulation the administration seeks to avoid.

With states passing their own AI laws and no federal framework in place, companies face a fragmented landscape. California requires AI firms to issue risk frameworks and report safety issues, while Florida has opened a criminal investigation into OpenAI over alleged consumer harms.

These details were first reported by CNN.