OpenAI delays GPT-5.6 to a government-approved limited preview at White House request

OpenAI confirmed it will stagger the rollout of GPT-5.6, releasing it initially only to federal government-approved enterprise customers during a preview period rather than shipping broadly. According to reports from The Verge, Axios and Politico, the request came from the Office of the National Cyber Director and the Office of Science and Technology Policy, and during the preview the administration would approve customer access on a case-by-case basis as it stands up a formal model security testing framework. Altman relayed the decision in an internal Q&A and memo, stressing this is a response to government directives and not OpenAI's preferred long-term model.

Mechanically, this is a sharp departure from OpenAI's usual launch playbook of broad API and ChatGPT availability on day one. A case-by-case federal approval gate inserts the government directly into the commercial deployment loop, turning a product launch into a regulated-access process. The timing is notable: OpenAI is simultaneously reported to be preparing a confidential IPO filing, with valuation chatter around $1 trillion and debate over whether the public debut lands in late 2026 or 2027.

Competitively, this is the second frontier lab in a month to face Washington-imposed release friction — Anthropic's Claude Fable 5 and Mythos 5 were restricted for foreign nationals in June. The asymmetry is becoming the story: GPT-5.5 remains unrestricted while GPT-5.6 is gated, and Anthropic's most powerful models are export-controlled. Developers and policy watchers warn this introduces a new operational risk — frontier-model availability now depends on political relationships, not just technical readiness.

Watch whether other labs (Google, Meta, xAI) get similar requests, how 'government-approved customer' is defined, and whether the preview framework becomes a durable precedent for how the most capable US models reach the market.