OpenAI Limits GPT-5.6 Preview After White House Request

OpenAI Puts GPT-5.6 Behind A Trusted-Partner Preview

OpenAI will make its GPT-5.6 Sol, Terra and Luna models available first through a limited preview for a small group of trusted partners, after a request from the White House.

The company said it had shown the models and their capabilities to the US federal government before the wider release.

Participants in the preview are partners whose participation has been shared with the government.

The decision changes the first step of a major model launch.

OpenAI is not describing GPT-5.6 as a normal public release at the outset, and it is not opening the preview to every developer, company or user that might want early access.

OpenAI said the restriction is a short-term step intended to support broader availability in the coming weeks.

The company also said it is working with the administration on a cyber Executive Order framework and a repeatable process for future model releases.

Model Access Becomes A Policy Control

The White House request puts AI model distribution inside a policy process before the full commercial rollout.

For enterprise customers and developers, the immediate issue is access: OpenAI has named the models, but the first preview depends on a trusted-partner list that has been shared with the government.

The move follows a separate confrontation involving Anthropic.

Earlier in June, the Department of Commerce forced Anthropic to disable access to its Fable 5 and Mythos 5 models for any foreign national, whether inside or outside the US.

Anthropic said the Commerce Department cited a vague risk that the models could be compromised.

The company said it had not received disclosure of a concerning non-universal jailbreak that led to a harmful result.

A group of at least 100 AI executives, cybersecurity experts and academics raised objections in a letter addressed to Commerce Secretary Howard Lutnick and Cyber Director Sean Cairncross.

Their concern was that the administration's approach could give a small number of officials control over a powerful expressive technology without enough transparency or accountability.

Release Timing Still Depends On The Framework

The policy tension is sharper because President Donald Trump has argued for a less regulatory and more innovation-centered approach to AI.

The latest OpenAI and Anthropic restrictions show that advanced-model access can still become a government-control issue when officials connect model capability to security risk.

The State Department defended the broader approach by saying American technology dominance and protection of critical technologies can move together.

It described the Pax Silica coalition as built around that same balance between leadership and controls.

For companies that build on frontier models, the practical effect is narrower than a full ban but still material.

Early access may now depend not only on commercial readiness or safety testing, but also on whether a partner is acceptable inside a government-reviewed release process.

OpenAI has not named the trusted partners, the exact date for broader GPT-5.6 availability or the release rules that will apply after the cyber Executive Order framework is agreed.

The launch now depends on whether OpenAI and the administration can turn a one-off preview limit into a repeatable process for future model releases.