The U.S. government has inserted itself directly into how frontier AI models reach the public, requiring both OpenAI and Anthropic to submit their newest systems for federal approval before broader release. OpenAI announced Friday that its new GPT-5.6 Sol model will be available only to customers approved by the Trump administration, while Anthropic received partial clearance to deploy its Mythos 5 model to a limited group of cyber defenders and infrastructure providers.

What Triggered This Government Gatekeeping?

The regulatory intervention stems from concerns about AI's cybersecurity capabilities. Earlier this year, Anthropic warned that its Mythos model could identify software vulnerabilities in ways that malicious hackers might weaponize, potentially threatening critical computer networks. This disclosure alarmed White House officials and prompted Trump to sign an executive order in June establishing a framework for federal vetting of advanced AI systems for up to 30 days before public release.

David Sacks, who co-leads Trump's council of technology and science advisers, characterized the situation bluntly. "Dario came to Washington a few months ago, back in April, and basically said that he had created a cyber weapon called Mythos," Sacks said on a recent podcast, referring to Anthropic CEO Dario Amodei. The disclosure, whether intentional or not, elevated AI safety concerns from academic discussion to national security priority.

Anthropic had initially pulled two models, Fable 5 and Mythos 5, from broader release just days after unveiling them to comply with a Trump directive blocking their use by foreign nationals. Two weeks later, the government lifted restrictions on Mythos 5, but Fable remains offline with no clear timeline for public availability.

Why Does This Approval Process Matter for AI Companies?

The government's case-by-case approval system creates economic pressure on AI labs. Frontier models are extraordinarily expensive to train, and companies depend on broad customer access to recoup those costs quickly. Every week a model sits in regulatory review represents lost revenue and delays the return on massive infrastructure investments.

OpenAI has not publicly named the roughly 20 customers approved to use GPT-5.6 Sol so far, and CEO Sam Altman reportedly projected the preview period at a couple of weeks. Anthropic's Mythos, by contrast, has already been in limited preview for months with no general-release date announced, suggesting the approval process could stretch far longer than initially expected.

The timing compounds the challenge. Both companies are exploring paths to go public on Wall Street, following SpaceX's record-setting initial public offering in June. Unpredictable government intervention in product releases adds uncertainty that investors scrutinize closely, especially when the data center buildout financing the entire AI industry depends on consistent revenue from new model deployments.

What Are the Structural Problems With This Approval System?

Experts and industry observers have identified critical gaps in how the government is conducting these reviews. The federal government lacks in-house expertise and capacity to evaluate frontier AI models properly, and regulators have not publicly articulated what specific risks they are trying to prevent. This creates a situation where officials have authority to block releases but no shared definition of what they are blocking against.

"I just want to say that pretty much nobody in the cybersecurity industry believes that there's any factual basis for this action," said Alex Stamos, a Stanford University cybersecurity expert and chief product officer at AI security company Corridor.

Alex Stamos, Chief Product Officer at Corridor and former Chief Security Officer at Meta

Stamos reviewed Amazon's analysis of Anthropic's Fable model and found no risks that aren't already present in other publicly available AI models, including those made in China. He argued that restricting U.S. AI releases while competitors abroad operate freely undermines American competitiveness. "If the administration is honest about wanting the United States to beat China in this race, then this is about the dumbest thing they could possibly do," Stamos stated.

U.S. Representative Lori Trahan, a Massachusetts Democrat and co-author of a bipartisan AI regulation bill, expressed concern about the ad-hoc nature of the process. "The Trump administration is deciding company by company who gets access to the newest AI model. No law. No process. No oversight. Just appointees in Washington deciding who's in and who's out," Trahan said in a statement.

How Should the AI Industry Navigate This New Reality?

Industry observers argue that OpenAI and Anthropic now face the same approval bottleneck and should stop treating regulation as a competitive advantage. Dean Ball, a fellow at George Mason University and incoming OpenAI employee, outlined a collaborative approach in a post published the morning the GPT-5.6 news broke.

"It will mean lining up behind the least-bad regulatory options available, instead of fighting every regulation tooth and nail. And most of all, it will mean fighting for AI as an industry, instead of seeing safety and regulation as opportunities to gain an advantage," Ball explained.

Dean Ball, Fellow at George Mason University and Incoming Employee at OpenAI

Ball's prescription involves three key steps for the industry to move forward effectively:

• Define Clear Standards: AI labs should work with independent groups to establish what safety assurances would actually satisfy regulators, rather than operating under vague threat models that invite endless scrutiny.
• Accept Collective Responsibility: Companies must trust independent auditors and evaluators even when their priorities don't fully align with any single lab's interests, recognizing that the approval process now affects all competitors equally.
• Coordinate on Evaluation Criteria: The industry should help the government define what to test for, because a regulator without clear guidance will test for everything slowly, delaying all releases across the board.

The underlying risks regulators are addressing are real. AI tools are reshaping cybersecurity workflows with measurable consequences for both offense and defense, and similar dynamics are playing out in biological risk assessment and AI alignment research. However, simply restricting model releases won't address these risks on its own; it mostly restricts what reaches the public while competitors abroad continue developing similar capabilities.

What Happens Next for OpenAI and Anthropic?

OpenAI said it does not believe government access processes should become the long-term default and views the testing period as temporary, with broader availability expected in the coming weeks. The company acknowledged that GPT-5.6 Sol "is better at helping people find and fix vulnerabilities" than at carrying out cyberattacks and does not cross its own risk threshold, but uncertainty about unforeseen risks, especially if combined with other tools, justified the phased approach.

Anthropic expressed being "pleased" by the partial release of Mythos and said it will "continue to work with the government to expand access" and make Fable available again to general users. Commerce Secretary Howard Lutnick told Anthropic in a letter that its work to address government concerns "yielded significant progress," signaling that continued cooperation could lead to broader clearance.

The stakes extend beyond these two companies. The labs that figure out how to negotiate the approval queue collectively, establishing evaluation standards and identifying which rules are worth absorbing, will set the pace of U.S. AI deployment for the next several years. Those that continue treating regulation as a competitive wedge will discover that the wedge cuts both ways, and that the data center financing the entire industry is watching release cadence very closely.