Anthropic dispute tests future of frontier AI access

Anthropic executives met with Trump administration officials in Washington on Monday in an effort to reverse a government directive that led the company last week to suspend access to its newest artificial intelligence models.

The meetings followed an export control directive issued on Friday that required Anthropic to restrict access to its newly released Fable 5 and Mythos 5 models for foreign nationals over unspecified national security concerns. The dispute has fueled debate among researchers and industry executives over whether the most advanced AI models will face greater government scrutiny and whether companies that build on them can count on uninterrupted access.

Anthropic received the directive days after unveiling Fable 5 and Mythos 5, which the company described as state-of-the-art models with advanced cybersecurity capabilities. Anthropic made Fable 5 broadly available with additional safeguards, while limiting access to Mythos 5, a less-restricted version of the model, to participants in Project Glasswing, a program that provides select cybersecurity organizations and operators of critical infrastructure with access to the company’s most advanced AI systems. IBM announced in May that it participates in Project Glasswing.

The order required Anthropic to restrict access to the models for foreign nationals. Anthropic responded by disabling the systems for all customers while it worked to comply.

CNBC reported that government officials cited an unspecified national security threat when directing Anthropic to take the models offline.

Anthropic has said the government’s concerns involve a possible way for users to get around some of the model’s cybersecurity protections.

The Anthropic dispute illustrates how government action can quickly affect access to commercially available AI systems, Kush Varshney, an IBM Fellow and AI researcher, told IBM Think in an interview. “This moment shows us that the age of greater government restrictions on models is here,” he added.

Restrictions tied to advanced AI systems could also affect how companies recruit and compete for talent, Gabe Goodhart, Chief Architect of AI Open Innovation at IBM, told IBM Think in an interview.

“Enterprises need to be nimble and expect volatility from model providers,” Goodhart said. “It also means that enterprises need to factor in the possibility of sudden regulatory action towards their AI products.”

Goodhart said the directive’s treatment of foreign nationals could have unintended consequences for the US AI industry.

“The particularly worrying thing is that the executive order prohibits even Anthropic employees who are foreign nationals from using the model,” he said. “Having worked in defense before, the huge overhead of employing foreign citizens meant that, in practice, these companies almost exclusively employed American citizens and therefore had a significant disadvantage in the overall talent pool. I worry that this could be a big blow for American AI if US labs are effectively firewalled from top international talent.”