Trump Delays OpenAI: Launch of GPT-5.6 Postponed

Trump Slows OpenAI: Launch of GPT-5.6 Delayed

In an unexpected move, the White House has requested OpenAI to delay the launch of its GPT-5.6 model, a decision that has surprised the tech industry. This request, revealed by The Information and Reuters, is particularly surprising given that the Trump administration had previously committed to a deregulatory policy regarding artificial intelligence. Indeed, Donald Trump had made economic freedom a cornerstone of his policy, dismantling several decrees from the Biden era and removing regulatory safeguards, which had been welcomed by Silicon Valley. However, the reason given for this delay is cybersecurity, a crucial issue that has led to temporarily limiting access to the model to a select group of partners, under the direct supervision of the government. An internal memo from Sam Altman, CEO of OpenAI, confirms this initial restriction, stating that access to the model would be limited to a few carefully chosen partners, with client-by-client validation by the government.

An Intervention from Little-Known Offices

The request for a delay did not come directly from Donald Trump, but from two rarely spotlighted U.S. executive offices: the Office of the National Cyber Director and the Office of Science and Technology Policy. These entities expressed concerns about the capabilities of the GPT-5.6 model, fearing it could be exploited by malicious actors to conduct cyberattacks. Discussions began during a meeting between Sam Altman and government representatives, who were keen that all relevant administrations had time to test the model. Rather than a total ban, the administration opted for a phased deployment, initially limiting access to selected partners before a broader rollout. In his memo, Altman emphasized that this framework goes against their "preferred long term model," a procedure he accepts but does not wish to see become permanent.

A Decree in the Background

This intervention is based on a decree signed on June 2, 2026, titled "Promoting Advanced AI Innovation and Security." This document mandates a security review before the release of advanced AI models, granting government teams a thirty-day period to evaluate these models. Although this review is presented as optional, the request from the two White House offices makes this step nearly mandatory. This highlights the tension between the desire to promote innovation and the need to ensure security.

A Two-Part Precedent

OpenAI is no stranger to phased deployments. In 2019, the company had already delayed the full release of GPT-2 for nine months due to concerns about potential misuse. More recently, GPT-5.5 was initially reserved for paying subscribers before becoming accessible to free users. A "cyber" version of GPT-5.5 was also distributed in restricted access to accredited defenders. However, the current situation differs in that it is now Washington controlling the pace of dissemination, validating each access individually. This approach marks a new stage in the regulation of AI technologies, where the federal government plays a direct role in managing access to advanced models.

A few weeks prior, the most advanced models from Anthropic, including Claude Mythos 5 and Fable 5, also saw their access restricted. This time, it was through export controls to prevent their use by foreign nationals. Authorities justified their intervention on GPT-5.6 by citing capabilities deemed comparable to those of Mythos, which prompted the White House's urgency to intervene. With OpenAI, however, the precedent crosses an additional threshold. It is no longer about restricting access to a model abroad, but delaying its release on American soil itself, for American partners, under the pretext of an evaluation that has produced no public results to date.

Implications for Europe

For European developers using OpenAI's APIs, this regulation means that the latest technological capabilities will be available later. European companies must wait for the circle of partners authorized by the U.S. government to expand. This situation highlights the differing approaches between the United States and Europe regarding AI regulation. While Europe is moving forward with clear legislative texts like the AI Act, the U.S. opts for a more flexible and informal regulation, without a strict legal framework. This divergence fuels the debate on technological sovereignty, strengthening arguments for an independent European AI.