White House Tells OpenAI: GPT-5.6? Sorry, Sol, Terra and Luna Only on Gov't Guest List 🏛️

OpenAI on Friday launched GPT-5.6 under a limited preview, releasing its Sol, Terra, and Luna models to a small group of government-approved partners at the request of the Trump administration while federal officials evaluate the system. The company said it previewed the models with the U.S. government before launch and is starting with restricted access as the two sides develop a process for future frontier AI releases. "As part of our ongoing engagement with the U.S. government, we previewed our plans and the models' capabilities ahead of today's launch," OpenAI wrote. "At their request, we are starting with a limited preview for a small group of trusted partners whose participation has been shared with the government, before releasing more broadly."

The administration's request, reported by The Information and Axios, was driven by what sources described as GPT-5.6's "Mythos-like" capabilities rather than a broader shift in AI policy. It came from the White House's Office of the National Cyber Director and Office of Science and Technology Policy, which asked OpenAI to limit the rollout while officials build a framework for evaluating advanced AI models before wider deployment. OpenAI CEO Sam Altman told employees in a memo on Thursday that the government is giving the company "access" through an approval system, adding that the arrangement does not represent OpenAI's long-term strategy.

GPT-5.6 Sol is positioned as the flagship, with OpenAI stating it improves performance across coding, biology, and cybersecurity. The company said Terra delivers performance comparable to GPT-5.5 at lower cost, and Luna is built for high-volume, low-cost workloads, with "max" and "ultra" reasoning modes giving Sol more time to solve complex problems or coordinate multiple subagents. OpenAI claimed Sol achieved the highest score on TerminalBench, a command-line software engineering benchmark, and outperformed GPT-5.5, Claude Mythos 5, and Fable 5, while posting stronger results than GPT-5.5 on GeneBench v1 using fewer tokens.

On cybersecurity, OpenAI said GPT-5.6 pairs stronger capabilities with expanded safeguards for defensive research while limiting offensive misuse, and that the model remains below the company's Cyber Critical threshold because it could not autonomously produce a complete exploit chain during testing. The release follows the administration's earlier order that Anthropic limit access to its Fable 5 and Mythos 5 models, after which Anthropic pulled the variants following concerns in Washington and on Wall Street over cyber capabilities and safety risks. President Trump's executive order earlier this month directed federal agencies to establish a voluntary testing framework for advanced AI systems before release, the framework under which GPT-5.6 is now being evaluated.

Anthropic, OpenAI, and Google have each published governance proposals calling for structured evaluations of the most capable models, greater transparency around safety testing, independent review of high-risk systems, and a larger government role in overseeing AI development. During Senate testimony in 2023, Altman urged lawmakers to establish a regulatory agency for advanced AI systems, while Anthropic CEO Dario Amodei has argued that the most capable models should undergo rigorous government-backed evaluations before deployment. Anthropic is also calling on Congress to strengthen protections against AI model distillation after alleging that operators affiliated with Alibaba and its Qwen AI lab generated more than 28.8 million exchanges with Claude between April and June, a claim tied to a June 10 letter from Anthropic to Senate Banking Committee Chairman Tim Scott and Ranking Member Elizabeth Warren.