The Trump administration partially lifts the ban on Mythos 5, OpenAI GPT-5.6 Sol is only available to White House-approved clients.

On Friday (6/27), the Trump administration partially lifted the ban on Anthropic's strongest cybersecurity model, Mythos 5, allowing it to be provided to a small group of network defenders and infrastructure providers; at the same time, OpenAI announced that GPT-5.6 Sol would only be available to clients approved by the White House, approximately 20 companies. The advanced models of the two AI giants are simultaneously subject to government review, marking a new phase in U.S. AI national security regulation.

(Previous summary: GPT-5.5-Cyber cybersecurity capabilities beat Claude Mythos! Two fates: White House clearance vs. being blocked)

(Background: OpenAI's new model GPT-5.6 not allowed: Trump administration demands phased release)

Table of Contents

Toggle

• The Mythos Incident: From "Cyber Weapon" to Partial Unban
• OpenAI's Low-Key Cooperation: Phased Release of GPT-5.6 Sol
• Cybersecurity Backlash: Stanford Expert Says "Completely Groundless"
• Regulatory Warming Under IPO Shadow

On Friday (6/27), the Trump administration dropped two bombshells on the AI industry: on one hand, it partially lifted the ban on Anthropic's cybersecurity model Mythos 5, which had been effectively blocked by the Commerce Department two weeks ago, allowing its redeployment to "a small group of network defenders and infrastructure providers"; on the other hand, OpenAI proactively complied with White House requirements, restricting the release of its latest flagship model GPT-5.6 Sol to about 20 clients approved by the Trump administration.

In its announcement, Anthropic expressed "welcome" for the partial release of Mythos 5 and said it would "continue to work with the government to expand usage permissions," while also striving to bring back online Fable 5, positioned as a "safer version." However, Fable 5 has been taken down for two weeks, and even with the partial unban of Mythos 5, Fable 5 has not yet been cleared.

The Mythos Incident: From "Cyber Weapon" to Partial Unban

The starting point of this AI regulatory storm was a warning from Anthropic CEO Dario Amodei, who visited Washington in April this year. In a briefing, he admitted that the Mythos model's ability to discover software vulnerabilities had reached a level where it "could be weaponized by malicious hackers," potentially threatening critical computer networks worldwide.

Trump's technology advisor and well-known investor David Sacks described the scene in a recent podcast: "Dario came to Washington a few months ago and basically said he had built a cyber weapon called Mythos. He raised everyone's cortisol levels through the roof, made everyone very nervous, and there's some truth to that—the model's cyber capabilities have indeed improved significantly."

In early June, Trump signed an AI regulatory executive order establishing a framework for the federal government to conduct national security risk reviews of the most advanced AI systems for up to 30 days. Subsequently, the Commerce Department effectively blocked Anthropic's Fable 5 and Mythos 5, prohibiting foreign nationals from using them. Anthropic was forced to take the models offline within days of their debut.

OpenAI's Low-Key Cooperation: Phased Release of GPT-5.6 Sol

Unlike Anthropic's public confrontation, OpenAI chose to cooperate quietly with the White House. The company announced on Friday that GPT-5.6 Sol would only be released to clients approved by the Trump administration, emphasizing that this was a transitional measure.

"We do not believe that such government review processes should become the long-term default," OpenAI said in its announcement, describing this testing period as "a temporary step on the path toward broader release in the coming weeks."

OpenAI emphasized that the Sol model is "better at helping people discover and fix vulnerabilities" rather than launching cyber attacks, and has not crossed the company's own risk threshold. However, OpenAI also acknowledged that combining the model with other tools could pose unforeseen risks. The company has not yet disclosed the list of the approximately 20 approved clients.

OpenAI CEO Sam Altman communicated with Commerce Secretary Howard Lutnick on Wednesday regarding the model release, part of a series of negotiations between AI industry leaders and Trump officials in recent weeks.

Cybersecurity Backlash: Stanford Expert Says "Completely Groundless"

The government's tough regulatory actions sparked strong backlash from the cybersecurity community. Alex Stamos, a cybersecurity expert at Stanford University, stated bluntly in a press call: "I just want to say that almost no one in the cybersecurity industry thinks this action has any factual basis."

Stamos, currently Chief Product Officer at AI security company Corridor and former Chief Security Officer of Facebook parent company Meta, said he had reviewed Amazon's analysis of Fable research—Amazon being Anthropic's primary cloud partner—and found no risks "that are not also present in other publicly available AI models, including those made in China."

"If the government is serious about beating China, then this is probably the dumbest thing they