Breaking

OpenAI to Stagger GPT-5.6 Release Under Government Approval

June 25, 2026 at 19:29 EDT

Foundation Models
Policy & Regulation

At the Trump administration's request, OpenAI will reportedly hold back a general release of its next model, GPT-5.6, instead offering it through a limited preview in which the government approves access one customer at a time. CEO Sam Altman explained the plan to employees during an internal Q&A, citing security concerns as the reason.

June 2026 · Frontier AI Policy

Washington Becomes AI's Gatekeeper: GPT-5.6 Held to a Limited Preview

OpenAI will release its newest model only to a small set of trusted partners after the Trump administration asked it to stagger the rollout over security concerns — with the government approving access customer by customer.

~20

trusted partner firms in the preview (possibly via Amazon Bedrock)

1.5M

estimated context window, in tokens

5 of 6

major labs agreed to government review — Meta the lone holdout

Context window, drawn to scale

Each block ≈ 0.25M tokens · GPT-5.6 is reportedly ~50% larger than GPT-5.5

~1.0M

GPT-5.5 API

~1.5M

GPT-5.6

How access now flows

A White House framework's "voluntary pre-release cooperation" is reportedly becoming a de facto gatekeeping step.

OpenAI builds

GPT-5.6 ready

→

Gov't review

ONCD · OSTP

→

Approved one-by-one

~20 partners

The case for it

Preserves US lead while guarding against misuse of frontier models
Builds on cyber-defense wins from GPT-5.5-Cyber and Daybreak
1.5M context could transform large codebases and long documents

The concerns

Critics call it a de facto licensing regime with a "chilling effect"
A setback as China's GLM-5.2 ships low-cost, open-weight models
Broad availability timing undetermined; few real-world evals exist

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.