OpenAI Calls Government AI Vetting 'Voluntary' Then Says It Shouldn't Exist, Exposing A Compliance Squeeze

Sam Altman publicly blessed President Trump's AI executive order as striking exactly the right balance, then privately admitted to his own staff that he never expected it would let the government vet his customers one by one.

The OpenAI chief executive praised the order within hours of its signing on 2 June 2026. Three weeks later, his company was forced to stagger the launch of its most powerful model, GPT-5.6 Sol, while federal officials approved each early customer.

The whiplash between the public endorsement and the private alarm now sits at the centre of the most consequential clash yet between Washington and the AI industry.

The Endorsement That Came Back To Bite

Altman left no doubt about his support when the order landed. In a post on X on the evening of 2 June, he wrote that the United States should lead on AI by building the best models, keeping them safe and getting cyber tools to trusted defenders, before declaring that 'the new EO gets the balance right'.

He was not a bystander to the policy. OpenAI was one of the firms working with the White House as the directive took shape, and Altman spent the following days meeting administration officials and congressional leaders in Washington.

The order he welcomed set up what the White House repeatedly called a voluntary framework. It asked developers to give the government up to 30 days of early access to a 'covered frontier model' before wider release, and let officials help choose which trusted partners received it first.

On its face, the arrangement read as a light-touch security review rather than a gate on commerce. Trump himself had hesitated over it, pulling an earlier draft in May because he feared a longer review window would block AI development.

The Memo That Told A Different Story

The reality that arrived with GPT-5.6 Sol looked far heavier. According to an internal memo first reported by The Information, Altman told employees the administration would approve access customer by customer during the preview period.

OpenAI had expected it might need to stagger the release, yet did not anticipate restrictions as severe as per-customer approval and a cap of around 20 partners at launch. For a company whose ChatGPT service now claims roughly one billion monthly users, locking the flagship behind a federally approved guest list of 20 is a remarkable constraint.

That is a stark distance from 'gets the balance right'. The man who had vouched for the framework in public was, weeks later, explaining to his own workforce that it bit harder than he bargained for.

OpenAI dropped its objection into the open on 26 June, stating in its launch announcement that it does not believe this kind of government access process 'should become the long-term default,' because it keeps the best tools from the developers, enterprises, and cyber defenders who need them.

There is a charitable reading, and fairness demands stating it. The order's fine print was always going to be written later, through a classified benchmarking process and agency rules due by late July, so the squeeze Altman now feels may reflect a framework that hardened after he endorsed it rather than a bait and switch. Even so, a chief executive who shaped and praised a policy ended up surprised by its teeth.

A Cosy Relationship With A Sharp Edge

The episode is more striking because of how close OpenAI and the administration have grown. CNBC reported that the company and the White House have been in talks for more than a year about a possible government equity stake in OpenAI, an idea Altman first floated in 2025 and tied to a proposed public wealth fund. The same government that may one day own a slice of the company is now deciding who may buy its newest model.

OpenAI insists it still trusts the administration's intentions, saying it believes the White House has the interests of US AI competitiveness at heart. The company frames itself as caught in an unfinished process, cooperating while the rules are still being drawn. Behind that diplomacy, the commercial stakes are enormous for a firm valued at more than £671 billion ($850 billion) and preparing for a possible stock market debut.

The friction is no longer hypothetical. One AI legal startup has already sued the government, arguing that losing access to rival Anthropic's restricted Fable model inflicted immediate and existential harm, an early sign of the damage that vetted, staggered rollouts can do to businesses built on these tools.

Altman bet that endorsing the government's hand would keep it light, and GPT-5.6 Sol is the first measure of how badly that bet may have misjudged the grip.