Trump White House Forces Staggered GPT-5.6 Release with Customer-by-Customer Approval

The Trump administration — through the Office of the National Cyber Director and the Office of Science and Technology Policy — has requested that OpenAI delay and stagger the release of its upcoming GPT-5.6 model. Government agencies are now actively approving enterprise customer access on a customer-by-customer basis during the preview period.

OpenAI CEO Sam Altman confirmed the arrangement in an internal employee Q&A, calling it 'not our preferred long-term model' while signaling full cooperation.

Key details:

• GPT-5.6 is described by government officials as 'on par' with Anthropic's Mythos in advanced autonomous cybersecurity capabilities — specifically the ability to identify and exploit software vulnerabilities at levels that could breach critical infrastructure

• The mechanism is a voluntary 30-day pre-release review window established under a June executive order on advanced AI innovation and security

• Commerce Secretary Howard Lutnick and National Cyber Director advisors are directly approving specific enterprise customers for access

• A broader public release is expected 'a couple of weeks later' once government review is complete

• This closely mirrors OpenAI's gated deployment for GPT-5.4-Cyber, which was released only to vetted security teams

Compared to the Fable 5 situation — a sudden total ban with no advance notice — the GPT-5.6 stagger represents a more cooperative, commercially sustainable approach. OpenAI continues commercializing, enterprise customers receive access through a mediated process, and the government maintains oversight without cutting revenue entirely.

Why It Matters: This establishes a new template for frontier model governance. The era of immediate state-of-the-art model access ends — advanced capabilities will now be mediated by government relationships and security credentials rather than simple API subscriptions.