The big picture: OpenAI has launched the GPT-5.6 series in a limited preview following a directive from the Trump administration to delay the full public release over national security concerns. The new AI lineup includes the Sol, Terra, and Luna models that the company claims offer both general users and developers more control over their AI requirements, speed, and cost.
Sol, the flagship model in the GPT-5.6 lineup, is built with a robust safety stack with guardrails against higher-risk activities, sensitive cyber requests, and repeated misuse. Terra is designed for balanced reasoning and agentic workloads, with OpenAI claiming that it offers similar performance to GPT-5.5 while being 2x cheaper. Luna is the entry-level model targeted at budget-conscious users.
The Information reports that OpenAI CEO Sam Altman informed staff on Friday that the full public release of the new frontier AI models will be delayed. Altman noted that the federal government asked the company to stagger the launch to a small group of customers through a limited preview, subject to approval by the administration on a case-by-case basis.
The directive was reportedly issued following concerns raised by the Office of the National Cyber Director and the Office of Science and Technology Policy. The report adds that the administration believes the staggered release will allow US intelligence agencies to identify potential threats before malicious actors and rogue foreign governments can access the new technology.
– OpenAI (@OpenAI) June 26, 2026
It is unclear what specific eligibility criteria the Trump administration has set for potential customers to access OpenAI's latest frontier models. However, government officials are reportedly concerned about cyberattacks on critical public and military infrastructure, as well as misuse by military and mercenary forces controlled by rogue governments and warlords.
The directive comes a few weeks after President Donald Trump signed an executive order asking AI companies to get their frontier AI models reviewed by the federal government before their public release. Currently, there's no standardized framework to assess AI models for potential security risks, but the administration is expected to issue clarifications on the matter soon.
Altman described the decision to withhold the full public release of GPT-5.6 as a short-term step and said that complying with the executive order is the most effective path to broader availability in the coming weeks. He added that OpenAI is working with the administration to help shape the so-called "cyber Executive Order" framework, which is expected to provide clearer rules for future model releases.
These developments come just days after OpenAI rival Anthropic accused Chinese tech giant Alibaba of illegally extracting capabilities from its Claude AI model. The company also disabled public access to its frontier AI models "Mythos" and "Fable" over concerns that China and other blacklisted countries could use their advanced capabilities against US interests.