Cybersecurity Policy Report, At White House Request, OpenAI Delays Public Release of Cyber-Capable Models, (Jun 26, 2026)
OpenAI is withholding its newest frontier AI (artificial intelligence) models from public release—despite their potential to help cyber defenders—at the request of the Trump administration, the company said today.
OpenAI will offer a “limited preview” of its GPT–5.6 series, which includes Sol, Terra, and Luna, to “a small group of trusted partners whose participation has been shared with the government, before releasing more broadly,” the company said in a news release.
“We believe in broad access, and we plan to make GPT–5.6 Sol, Terra, and Luna generally available in the coming weeks. As part of our ongoing engagement with the U.S. government, we previewed our plans and the models’ capabilities ahead of today’s launch,” it said.
“We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them. We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the Administration,” it noted.
The Trump administration recently imposed export controls on a frontier model developed by Anthropic to prevent its use by foreigners. Unable to determine which users should be prevented from using the model, Anthropic canceled the public release of its Fable 5 and Mythos 5 models, which are based on Anthropic’s Claude Mythos2 Preview.
Mythos2 Preview proved to be so adept at finding and exploiting flaws in widely used software that Anthropic withheld it from public release and stripped those abilities from the public models. The White House nevertheless had concerns about the public models being used by malicious hackers (CPR, June 15).
The White House didn’t publicly comment on its concerns about the new OpenAI models. OpenAI said the models were good at finding vulnerabilities for cyber defenders but had robust guardrails to prevent them from being used by hackers.
“GPT–5.6 Sol is better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks. As these capabilities continue to advance, our priority is to make sure they reach and benefit defenders, who can use these tools to find weaknesses, develop patches, and strengthen systems more broadly,” OpenAI said.
“Our goal is to make prohibited offensive activity more difficult, uncertain, and detectable without unnecessarily limiting those beneficial uses,” the company added. “Based on our assessment of the model and safeguards, we expect substantial benefit for legitimate defensive work, while meaningfully constraining prohibited offensive use.”
MainStory: TopStory FederalLegislation DataSecurity AINews