OpenAI Delays GPT-5.6 Launch at U.S. Government Request

OpenAI announced three new AI models on Friday but will withhold them from general release following a request from the U.S. government, marking a significant shift in how frontier AI systems reach the market.

The company revealed GPT-5.6 Sol, Terra and Luna—named according to capability tiers, with Sol representing the most powerful offering. However, only a small group of undisclosed trusted partners will gain initial access while OpenAI negotiates broader availability with federal officials.

Government oversight expands

The restricted launch follows a Trump administration executive order signed earlier in June that asked AI developers to voluntarily submit models for government assessment before public release. While the order contained few specifics, its impact is already reshaping deployment timelines across the industry.

Two weeks before OpenAI's announcement, competitor Anthropic disabled access to two of its latest models to comply with an export control directive from the Trump administration. Those models remain offline while Anthropic continues negotiations with Washington officials.

OpenAI stated it previewed the new models' capabilities and shared deployment plans with the government ahead of Friday's announcement. The company emphasized it is working with the administration to establish an assessment framework and develop repeatable processes for future releases.

Why it matters

This represents the first major instance of a leading AI company delaying a model launch at government request under the new voluntary framework. The precedent could fundamentally alter how quickly advanced AI capabilities reach developers, enterprises and researchers—potentially slowing innovation cycles while raising questions about which entities gain early access and on what criteria.

Technical capabilities and safety assessments

GPT-5.6 Sol demonstrates improvements in coding and biology tasks, with OpenAI describing it as the company's most capable cybersecurity model to date. The company assessed that Sol performs better at helping users identify and fix vulnerabilities than executing complete attacks, and does not cross OpenAI's "critical" cybersecurity risk threshold—defined as creating unprecedented pathways to severe harm.

OpenAI expressed concern about the long-term implications of government-gated releases. "We don't believe this kind of government access process should become the long-term default," the company stated. "It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."

The company said it expects to make the models generally available within weeks, though no specific timeline was provided.

These details were first reported by CNBC.