OpenAI Limits GPT-5.6 to Trusted Partners

What did OpenAI just release?

OpenAI announced three new AI models on June 26, 2026: GPT-5.6 Sol, Terra, and Luna. Sol is the most advanced of the three. Terra sits in the middle tier. Luna is built for lower-cost use cases.

Rather than a full public launch, OpenAI restricted access to a small group of vetted "trusted partners" — a decision made at the Trump administration's request, according to Quartz.

Why is access limited to trusted partners?

The Trump administration asked OpenAI to slow the rollout over national security and safety concerns. Specifically, the Office of the National Cyber Director and the Office of Science and Technology Policy requested the limited release, TechCrunch reported.

OpenAI CEO Sam Altman reportedly told staff that the government would be "approving access customer by customer" during the preview period.

Who are the 20 trusted partners?

Access is being extended to 20 partners. OpenAI did not disclose the names of those partners. Amazon's Bedrock software platform is confirmed as one route in, according to Quartz.

OpenAI previewed the models' capabilities with government officials before the June 26 launch.

What are the cybersecurity risk levels for these models?

OpenAI placed all three models in the high-capability category for cybersecurity and biological and chemical risk under its Preparedness Framework. Here is how the company characterized the risks:

You might also like

Identity Change Is the Hardest Code to Ship

Authentic Beats Perfect in the AI Tsunami

Why People Resist AI (it's a Standards Problem)

How I'm Rebuilding YouTube Trust After 3 Niche Pivots

OpenAI said none of the three models cross its internal "critical" cybersecurity risk threshold. It also noted that novel dangers could surface when the models are used alongside other technologies.

What did OpenAI say about this process becoming permanent?

OpenAI was direct: it does not want this to be the long-term standard.

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

The company said its near-term goal is to establish a consistent set of procedures — alongside the administration — for how future models are assessed and released.

What executive order is behind this?

President Trump signed an executive order earlier in June creating a voluntary framework for AI developers. Under that order, companies can submit "covered frontier models" to the U.S. government for up to 30 days before releasing them to trusted partners. The rules underpinning that process have not yet been finalized, according to The Wall Street Journal, as cited by Quartz.

OpenAI said it is taking this step as it works with the administration to develop the framework and "a repeatable process for future model releases."

How does this connect to what happened with Anthropic?

This rollout follows a turbulent episode involving Anthropic. Earlier in June, the Trump administration barred foreign nationals from accessing Anthropic's Mythos 5 and Fable 5 models, citing concerns about cyberattack misuse. Enforcing a nationality-based restriction proved technically impossible. Anthropic had to pull both models entirely — after receiving roughly 90 minutes of warning.

That episode is part of why Washington is now pressing for earlier access to frontier AI model safety reviews before public launches happen.

The concern with powerful cyber-focused models is specific: they can identify and exploit software vulnerabilities at speeds no human analyst could match. This is the same issue that drove Anthropic to limit access to its Claude Mythos model through a program called Project Glasswing.

What is the broader context for AI oversight in 2026?

The Trump administration originally positioned itself as taking a "hands-off" approach to AI. In recent months, it has moved toward federal oversight of new frontier models. This shift is happening alongside broader US AI jobs and investment policy debates in Washington.

Here is a timeline of key events:

• Earlier in 2026: Anthropic releases Claude Mythos only to select partners via Project Glasswing
• Early June 2026: Trump signs executive order on voluntary AI model review framework
• Mid-June 2026: Trump administration bars foreign nationals from Anthropic's Mythos 5 and Fable 5; Anthropic pulls both models
• June 25, 2026: Sam Altman tells OpenAI staff about the government's customer-by-customer approval plan
• June 26, 2026: OpenAI officially announces GPT-5.6 Sol, Terra, and Luna with restricted access

As we read the sources, the Anthropic episode appears to be the direct trigger that pushed OpenAI and the administration toward a more structured — if still informal — pre-release review process.

The growing concern in Washington is that powerful AI cybersecurity risk tools could be exploited for malicious attacks before companies or governments can respond. LLMs have proven capable of writing malware, and some can execute ransomware attacks autonomously, according to TechCrunch.

When will GPT-5.6 be available to the public?

OpenAI said it plans to make GPT-5.6 Sol, Terra, and Luna broadly available in the coming weeks. Altman reportedly told staff that if the limited release goes well, a general release could follow "a couple of weeks later."

The company said it will continue testing and close coordination with partners as it prepares for the wider release. The government's Center for AI Standards and Innovation had the models under review, according to The Wall Street Journal as cited by Quartz. Commerce Secretary Howard Lutnick and Sam Altman discussed the GPT-5.6 launch on Wednesday, June 25.