The most powerful commercially available AI model in the world was cleared for mass public release this week through a process that involved no independent safety review, no finalized regulatory standards, and no public accountability mechanism. The U.S. Department of Commerce approved OpenAI's GPT-5.6 — including its flagship model Sol and lower tiers Terra and Luna — following private meetings between company technical staff and government officials, Axios first reported. The models go live Thursday.
What the Axios report describes as a resolution is actually a template — and the template is: powerful AI companies negotiate access to the public directly with the government, case by case, in private, without the structural guardrails that the administration's own executive order says should exist but hasn't built yet.
The Commerce Department's Center for AI Standards and Innovation conducted the testing. OpenAI sent technical experts to Washington who, according to the source cited by Axios, remained available to answer questions. That arrangement — a company deploying its own engineers to assist the government office evaluating its product — is not independent review. It is something closer to a guided tour.
President Trump's most recent AI executive order calls for the development of concrete standards for releasing advanced AI models. As of this week's GPT-5.6 clearance, those standards have not been finalized. OpenAI itself acknowledged the gap, stating publicly that AI firms and the government are operating before more concrete release standards exist. The clearance happened anyway.
OpenAI has been candid about the awkwardness of this arrangement. The company said publicly that the staggered rollout the administration imposed last month — limiting initial GPT-5.6 access to government-approved entities — was not its preferred approach. OpenAI also acknowledged that both companies and regulators are moving before the formal standards called for in the executive order have been written. That admission is not a minor disclosure. It is a description of a regulatory vacuum being filled by bilateral negotiation between the government and the companies it is supposed to oversee.
OpenAI is not alone in this. Anthropic's Mythos and Fable models were both subjected to Commerce Department restrictions in June — the department banned foreign access entirely, effectively pulling both from international markets. The ban on Fable was lifted last week. Mythos restrictions remain in place. The pattern across both companies is the same: access decisions are being made through a process that has no published criteria, no independent verification, and no mechanism for public challenge.
The accountability question here is structural, not personal. This is not primarily a story about whether any individual official acted in bad faith. It is a story about what happens when a government decides that the entities capable of deploying the most consequential technology in modern history get to participate in designing the process that governs them — while the public does not.
Consider what a genuinely independent review process would require: published criteria for what constitutes acceptable risk in a frontier AI model, an evaluation body with no financial relationship to the companies being assessed, a public comment period, and a documented record of the findings. None of those elements were present in the GPT-5.6 clearance. What existed instead was a government office, an AI company's technical staff, and a series of meetings whose content has not been made public.
The global dimension of this deserves attention. The Commerce Department's June ban on foreign access to Anthropic's models was framed as a national security measure — keeping advanced AI capabilities out of the hands of adversaries. But the effect was also to restrict access for researchers, developers, and institutions in allied nations who had no national security nexus. When the ban on Fable was lifted, customer access was restored within a day. No explanation of what changed was offered publicly. The criteria for who gets access to the world's most powerful AI tools, and when, is being decided in Washington through a process that the rest of the world cannot see or contest. As Tinsel News has previously reported on the G7's AI governance failures, the absence of multilateral frameworks is not an oversight — it is a choice that concentrates power in the hands of a small number of governments and companies.
The money runs in two directions at once. OpenAI has an obvious interest in broad, fast deployment — its business model depends on it, and every week of restricted access is a week of lost revenue and competitive ground ceded to rivals. The government has an interest in being seen as exercising oversight without actually building the institutional capacity that genuine oversight requires. Both interests are served by the current arrangement. The public interest in knowing whether GPT-5.6 poses risks that justify concern is not served at all.
This dynamic is not new to AI. It is the same pattern that allowed social media platforms to scale to billions of users before any serious regulatory framework existed — and the consequences of that lag are still accumulating. The difference with frontier AI is the speed of capability growth and the breadth of potential impact. A social media platform that amplifies misinformation does harm at scale. An AI model with reasoning capabilities that exceed human performance in multiple domains, deployed without independent safety verification, operates in a different risk category entirely. The argument for moving fast and building standards later made a certain kind of sense when the technology was narrow and the risks were bounded. That argument is harder to sustain now.
OpenAI's own public statements acknowledge this. The company said AI firms and the government are operating before more concrete standards have been finalized. That is a remarkable admission to make at the moment of a major product launch — and it passed without significant scrutiny in the coverage of the clearance. It should not have. A company releasing its most powerful model to date, in the same breath, conceding that the framework for deciding whether that release is safe does not yet exist, is not a minor disclosure buried in context. It is the story.
The Tinsel News AI Regulation Tracker documents the widening gap between AI capability deployment and the regulatory structures meant to govern it. The GPT-5.6 clearance is the latest entry. What it establishes is a precedent: when the standards don't exist yet, the government and the company negotiate directly, in private, and the result is called oversight. The next frontier model release — from OpenAI, from Anthropic, from whoever comes next — will inherit this template. The question is not whether that process is adequate. The question is who has the power to change it, and whether they will use it before the next capability threshold is crossed.
The administration's own executive order calls for the standards that would make this process legitimate. Those standards remain unwritten. GPT-5.6 launches Thursday anyway. The gap between what the government says it will do and what it actually does when a powerful company is waiting for an answer is now documented — and the documentation is the clearance itself.
