Sam Altman just told OpenAI employees that the Trump administration will personally vet and approve every customer who gets access to GPT-5.6, the company’s next major AI model.
The revelation marks an extraordinary moment in AI governance: a sitting U.S. administration has effectively seized approval authority over a private company’s product release. This is not a regulatory framework or safety review process operating at arm’s length. This is direct, case-by-case government veto power over which enterprises can use a new AI system.
- Direct Government Gatekeeping: The Trump administration will individually approve each enterprise customer seeking access to GPT-5.6 during its preview period, inserting federal decision-making directly into a private product release pipeline.
- Unequal Terms Across Competitors: OpenAI reportedly secured more favorable preview conditions than rival Anthropic, suggesting the administration is using approval authority as a negotiating lever with different terms for different companies.
- No Statutory Basis: The arrangement operates through commercial leverage rather than formal legislation or regulatory authority, setting a precedent that could extend to every major AI lab in the United States.
According to The Information, Altman disclosed the arrangement during a company Q&A on Wednesday. OpenAI will release GPT-5.6 in limited preview form, available only to a small group of enterprise customers initially. But access during that preview period requires explicit approval from the Trump administration itself. The federal government will evaluate each customer request and decide who gets in.
The Trump administration’s stated rationale centers on security concerns. Officials reportedly worry that unrestricted access to GPT-5.6 could create vulnerabilities or enable misuse. Rather than waiting for a full public release, they asked OpenAI to stagger the rollout and submit to government screening during the preview phase. The broader context here matters: as Harvard’s Edmond and Lily Safra Center for Ethics documents in its analysis of America’s AI Action Plan, the current administration has consistently prioritized innovation leverage over formal regulatory caution — making this kind of direct commercial negotiation a predictable governance instrument.
What Makes This Different From Standard Security Review?
What makes this arrangement significant is not the existence of security review — most AI companies now expect some form of government scrutiny. It is the granularity and directness of the control. The administration is not asking OpenAI to conduct internal safety testing or submit documentation for approval. It is inserting itself into the customer approval workflow, becoming the gatekeeper for commercial access to a cutting-edge AI model.
The deal OpenAI negotiated appears more favorable than what the Trump administration imposed on rival Anthropic, according to The Information’s reporting. That distinction matters: it suggests the administration is using approval authority as a negotiating tool, with different terms for different companies. OpenAI’s cooperation — and Altman’s public acknowledgment of the arrangement — may have earned the company a less restrictive preview period than competitors faced. This dynamic, where proximity to power determines product access, has direct parallels to how nations now compete for digital dominance by controlling access to strategic data assets rather than regulating them through transparent legal frameworks.
• GPT-5.6 preview access requires case-by-case federal approval for each enterprise customer, with no published criteria or timeline for decisions
• The arrangement applies during the preview period across all sectors — financial services, healthcare, defense, and media organizations must all seek government sign-off
• OpenAI’s terms are reportedly more favorable than those imposed on Anthropic, indicating differentiated treatment across competing AI labs
Who Actually Gets Approved — and What Does That Reveal?
For enterprise customers waiting to deploy GPT-5.6, the implication is immediate and practical. They cannot simply purchase or activate the model when it becomes available. They must submit to government vetting, provide details about their intended use case, and wait for federal approval. A financial services firm, a healthcare company, a defense contractor, a media organization — all would need the Trump administration’s sign-off before they could access the system.
This arrangement also reveals a shift in how the U.S. government is approaching AI governance. Rather than passing new legislation or establishing formal regulatory bodies, the Trump administration is negotiating directly with AI companies and embedding government decision-making into product release pipelines. It is governance through commercial leverage, not statutory authority. The 2025 AI Index Report from Stanford HAI tracks the accelerating pace at which governments worldwide are attempting to shape AI deployment — but the U.S. model emerging here is notably more direct than the legislative approaches taken by the EU or the UK.
• Governance scholars distinguish between arms-length regulatory review — where an independent body applies published criteria — and embedded approval authority, where the executive branch becomes a direct participant in commercial transactions
• The current arrangement falls squarely in the second category: there is no published framework, no independent review body, and no formal appeals process for denied customers
• This model concentrates discretionary power in the executive branch over which industries and organizations gain early access to transformative technology, with significant implications for competitive markets
Is This the Beginning of a Federal AI Clearance System?
The precedent extends beyond OpenAI. If the administration can extract case-by-case approval authority from OpenAI, it can likely demand similar arrangements from other AI labs. Google, Meta, Anthropic, and smaller AI companies may face comparable pressure. Over time, this could create a de facto system where no major AI model reaches broad commercial availability without federal clearance for each customer segment.
For users and businesses, the practical effect is a slowdown in AI deployment. GPT-5.6 was presumably built and ready for release. Instead, it enters a holding pattern while government offices process customer applications. That delays competitive advantage for early adopters, extends the timeline for AI integration across industries, and creates bureaucratic friction around a technology the administration itself views as strategically important. The question of who bears the cost of that friction — and who benefits from controlling it — is precisely the kind of structural power question that privacy tech investors are now pricing into their risk models as government entanglement with AI platforms deepens.
Altman’s decision to disclose the arrangement in a company Q&A — rather than having it leak or emerge in regulatory filings — suggests OpenAI is treating the government partnership as a selling point rather than a constraint. The message to employees: we negotiated a favorable deal with the Trump administration, and we are complying transparently. That framing may help with government relations, but it also normalizes the idea that federal approval is now part of the AI product release cycle.
What Comes After the Preview Period?
The question now is whether this preview-period arrangement is temporary or permanent. Will GPT-5.6 eventually reach full public availability once the administration completes its vetting, or does case-by-case approval become the standard access model going forward? Altman’s comments to employees did not specify an endpoint for government screening. Research from Stanford HAI on AI governance pathways consistently finds that temporary emergency frameworks have a strong tendency to become permanent institutional structures once the administrative infrastructure supporting them is in place.
Watch for announcements about which enterprise customers receive approval first — and which sectors the Trump administration prioritizes or restricts. That approval pattern will reveal what the administration actually views as the security risk, and more importantly, which industries it intends to shape through preferential access to the most capable AI systems available.
