Regulatory easing: Over 100 institutions approved to use Anthropic's most powerful model.

Author: Bao Yilong, Yang Chen; Source: Wall Street View

A regulatory standoff between the U.S. government and AI company Anthropic has seen a breakthrough, but it has not been fully resolved.

On Friday, according to reports in the media, U.S. Commerce Secretary Lutnick sent a letter to Anthropic announcing the removal of export restrictions on the company’s Claude Mythos 5 model, allowing more than 100 companies and federal agencies that have been approved by the government to access the model.

The move marks the most significant easing since the two sides’ two-week standoff began. However, restrictions on another flagship model, Fable 5—also banned—were not included in this lifting, and the export ban remains in effect.

Lutnick said that since the ban took effect, the government and Anthropic have continued conducting high-intensity negotiations with nearly daily frequency, and have achieved “significant progress.” On Friday, in a letter to Anthropic’s Chief Compute Officer Tom Brown, he wrote:

I have confirmed that the relevant security safeguards are in place, allowing certain trustworthy partners to access the Claude Mythos 5 model.

Two weeks ago, the government imposed export controls on Mythos, leading to the suspension of the release of both Mythos 5 and Fable 5. At the time, companies such as Amazon warned that these models could be “jailbroken,” and thus used for malicious purposes.

On Friday as well, Anthropic’s main competitor, OpenAI, rolled out its latest model, GPT-5.6, to a small number of partners that have been approved by the U.S. government.

Origin of the Ban: Security Concerns and Regulatory Friction

On June 12, the U.S. government imposed export controls on Anthropic’s Fable 5 and Mythos 5 models, banning use by foreign nationals, citing concerns that vulnerabilities existed in AI safety guardrails.

Before that, government officials and Anthropic CEO Dario Amodei had a series of heated phone calls, questioning whether the company’s safety protections were as reliable as it claimed.

Previously, Amazon CEO Andy Jassy called senior officials including U.S. Treasury Secretary Bessent, saying that Amazon’s researchers bypassed the safety guardrails of Anthropic’s flagship model Fable 5 through specific prompt phrases, obtaining sensitive information that could be used for cyberattacks. The White House then convened an emergency meeting.

After the ban was issued, Anthropic immediately announced that it would suspend access to both models for all users, saying the move was intended to ensure compliance, while insisting that “this is a misunderstanding.”

Anthropic is currently preparing for an IPO, and its relationship with the U.S. government has faced sustained pressure this year.

Previously, the company refused to allow the U.S. military to use its AI models for domestic surveillance and fully autonomous weapon systems, after which the government placed it on a national security blacklist.

The Scope of the Release Covers More Than 100 Institutions; Foreign Employees Also Benefit

Under the new arrangement announced on Friday, more than 100 entities approved by the government and listed in an annex—including many Fortune 500 companies and federal agencies—will no longer need to apply for export licenses in order to access the Claude Mythos 5 model.

According to Reuters, citing a person familiar with the matter who requested anonymity, foreign employees of the above institutions and Anthropic’s own foreign employees are also covered by this exemption. Institutions not on the approved list must still comply with the existing restrictions.

In his letter, Lutnick pointed out that Anthropic “has committed to cooperating with the U.S. government on agreements, standards, and release-related matters regarding its models.”

For Fable 5, which was also banned, Friday’s letter said nothing about it. According to Politico, citing a source close to the negotiations, the government is moving toward lifting restrictions on Fable 5, but the specific timetable remains unclear.

Industry Impact Reaches OpenAI; Regulatory Framework Is Becoming the Norm

On Friday, Anthropic’s main competitor OpenAI also announced that it would release its latest model GPT-5.6 to a limited group of partners that have been approved by the government, on the same day as the partial lifting of Anthropic’s ban.

This coincidence reflects the broad pressure the White House is exerting on the entire frontier AI industry.

According to Politico, White House pressure directly led OpenAI to restrict the scope of release of its most advanced model this week.

Analysts say that the U.S. Department of Commerce’s action may signal the beginning of routine regulation of frontier AI models by the U.S. government.

While the heads of frontier labs are deeply concerned about potentially losing time in the fierce global AI race, in the short term they may find it difficult to avoid this regulatory reality.