OpenAI limits access to new models due to government security concerns

OpenAI just dropped three new AI models. You just can’t use them yet.

On June 26, the company announced GPT-5.6 Sol, Terra, and Luna, but instead of the usual fanfare followed by immediate public access, OpenAI is keeping things behind velvet ropes. Only a select group of government-approved “trusted partners” will get early access, with broader availability coming after a review period. The reason: the Trump administration asked them to.

The White House called first

Before OpenAI made its announcement, it previewed the models’ capabilities to government officials. Those officials, focused on cybersecurity and national security implications, then requested that the company limit who gets to play with the new toys.

An executive order from the Trump administration encourages a 30-day pre-release security review by AI companies before launching advanced models.

OpenAI isn’t the only company getting this treatment. Anthropic, one of OpenAI’s primary competitors, faced similar US government directives regarding its Mythos model.

OpenAI has publicly stated that this kind of government access process should not become the long-term default. Their argument is straightforward: routine federal vetting of AI releases would stifle innovation, slow down development cycles, and ultimately hurt users, developers, and enterprises that depend on rapid access to new tools.

A new reality for AI companies

The security concerns aren’t abstract. Advanced AI models have demonstrated capabilities in areas like code generation, vulnerability discovery, and persuasion that make cybersecurity professionals genuinely nervous.

OpenAI cooperated with the government’s request this time, previewing capabilities and accepting restricted access, but is simultaneously arguing that this shouldn’t become routine. The 30-day pre-release review framework from the executive order suggests the administration is building toward a repeatable process. Anthropic finding itself in the same position with Mythos reinforces that this is an industry-wide phenomenon.

What this means for investors

For anyone with money in the AI sector, this development introduces regulatory timing risk. If the federal government can effectively delay model launches through security reviews, then the competitive dynamics of the AI race change meaningfully. Companies that build robust internal security review processes will have an advantage over those that don’t.

For now, the restricted access to Sol, Terra, and Luna is temporary. OpenAI has indicated broader access will follow once the review period concludes.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.