US to Set Early Access Rules for OpenAI's GPT-5.6 Release
OpenAI is reportedly planning to stagger the release of its next frontier model, GPT-5.6, after the U.S. government raised security concerns. According to The Information, the Trump administration has asked OpenAI to phase the launch, with federal officials approving preview access for each customer individually during the early window. This approach builds on OpenAI's existing practices, such as the delayed release of GPT-2 in 2019 and the tiered rollout of GPT-5.5 in April 2025. The plan extends the trusted-access model used for a cyber-focused version of GPT-5.5, now involving Washington directly. The framework aligns with Executive Order 14409, signed by President Trump on June 2, which requests frontier model developers to provide up to 30 days of government access before public release. A classified benchmark led by the NSA will determine which models are classified as frontier, based on advanced cyber capabilities. A Treasury-run clearinghouse will hunt and patch software flaws. The policy is voluntary and avoids licensing regimes. OpenAI has not officially confirmed GPT-5.6, with earlier timelines slipping toward July. How early access is managed could set a precedent for future models from OpenAI and Anthropic.
Key facts
- US government asks OpenAI to phase GPT-5.6 release over security concerns
- Federal officials approve early preview access one customer at a time
- Plan follows Executive Order 14409 for 30-day government access to frontier models
- NSA-led benchmark classifies models based on advanced cyber capabilities
- OpenAI not confirmed GPT-5.6; earlier timeline slipped to July