OpenAI to stagger the launch of GPT 5.6 at the request of the Trump administration

By The Guardian · June 26, 2026.

OpenAI has announced the launch of its new GPT 5.6 series of models in a phased manner, beginning only with a small group of partners considered trusted and whose identity has been shared with the U.S. government. The decision was not voluntary: the company acted at the express request of the Trump administration, through two federal agencies —the Office of the National Cyber Director (ONCD) and the Office of Science and Technology Policy (OSTP)—, which requested this gradual rollout as a precondition for any broader deployment.

Sam Altman himself communicated the situation to company employees in an internal memo, obtained by The Information, in which he explained that the government would approve access «customer by customer» during this trial period, with a general release expected «a couple of weeks later» if the process went smoothly. Altman made clear that this situation is not the one the company would have preferred: «We have made clear to the U.S. government that this is not our preferred model in the long term, and we will work with them and with other industry actors to achieve a more sustainable approach for future launches».

On the official blog, OpenAI was even more explicit about its displeasure, noting that this kind of government access process keeps the best AI tools away from «users, developers, businesses, cyber defenders and global partners who need them». However, the company called this step the «most solid path» to achieving broader availability in the coming weeks, while it works with the White House on developing a verification and deployment framework for new models, as required by an executive order recently signed by President Donald Trump.

The parallel with Anthropic is notable. The rival company had been involved in a similar episode with its Mythos model: it first voluntarily delayed its mass launch and, later, the U.S. government ordered the company to prevent foreign citizens from accessing public versions of the model, given that it possesses advanced cyberattack capabilities. According to the United Kingdom's AI security body, Mythos represents «a step forward» relative to previous frontier models. The key difference is that Anthropic initially acted on its own initiative, whereas in OpenAI's case government intervention was explicit and direct from the outset.

The federal pressure was not limited to the two agencies mentioned. According to The Information, Howard Lutnick, Trump's Commerce Secretary, personally intervened by calling Sam Altman to demand additional approvals from other agencies, even opposing a limited launch. This level of high-ranking involvement underscores the degree of scrutiny to which the most advanced AI models are being subjected by the U.S. administration.

As for the technical characteristics of GPT 5.6, the model comes in three variants: Sol, described as the «most powerful in the suite» and «OpenAI's most powerful model to date»; Terra, with slightly lower performance but lower cost for users; and Luna, the lowest-cost version. OpenAI has specified that Sol does not exceed the «critical cyber threshold» defined in its internal framework for assessing dangerous AI capabilities, and that the model is «better at helping people find and fix vulnerabilities than at reliably carrying out end-to-end attacks». This distinction is relevant given the context of government concern over the offensive cybersecurity capabilities of the most advanced models.

Regarding geographic access, all the entities that will receive GPT 5.6 in this initial phase will be U.S.-based. Nonetheless, OpenAI indicated that it expects to bring on foreign partners the following week, and that employees of the benefiting companies who work from abroad in «compatible countries» —which include the United Kingdom and Australia— will also have access to the model.

This episode is part of a significant shift in stance by the White House on AI. This month, Trump signed an executive order to create a voluntary framework allowing the federal government to evaluate the most powerful new AI models before their market launch. The turn contrasts with the administration's initial rhetoric: just last year, Vice President JD Vance had warned that «excessive regulation of the AI sector could kill a transformative industry». The change in position reflects growing concern over the national security implications of the latest-generation models, especially regarding their capabilities in the cyberspace domain.

As sector context, the tension between the development speed of the major AI labs and governments' ability to assess and mitigate risks is one of the most relevant lines of friction at the moment. The Trump government's request to OpenAI —and the precedents with Anthropic— suggest that the U.S. administration is, de facto, building a pre-launch review system for the most advanced models, although for now under a voluntary umbrella. The open question is whether this informal scheme will become binding regulation, and what implications that would have for the competitiveness of U.S. AI companies against global players not subject to the same restrictions.