OpenAI, the company behind ChatGPT, announced Friday that it is holding back its newest artificial intelligence model at the request of the Trump administration — marking the latest step in an unprecedented government review of AI technology over potential cybersecurity threats.
The new AI product, called GPT-5.6 Sol, will be made available only to a “small group of trusted partners” that have received approval from the Trump administration, OpenAI said.
In a statement, OpenAI made clear it has reservations about the arrangement going forward: “We don’t believe this kind of government access process should become the long-term default.” The company described the restricted rollout as a temporary measure on the “path to broader availability in the coming weeks.”
This cautious, phased launch comes on the heels of government action earlier this month against Anthropic, a competing AI company and maker of the Claude chatbot. Anthropic was forced to pull two of its newly released models — known as Fable 5 and Mythos 5 — offline just days after their public debut, in response to a Trump directive that blocked foreign nationals from accessing them.
Government officials have been on high alert since Anthropic flagged earlier this year that its Mythos model showed an unusual ability to identify weaknesses in software — a capability that could potentially be exploited by malicious hackers to attack critical computer systems worldwide.
President Trump signed an executive order on AI oversight in June, creating a framework that allows the federal government to evaluate the national security risks of the most powerful AI systems for up to 30 days before they become publicly available. While the order describes AI developer participation as voluntary, the full framework has not yet been put in place.
OpenAI said its Sol model “is better at helping people find and fix vulnerabilities” than it is at enabling cyberattacks, and that it does not exceed the company’s own internal risk limits. However, OpenAI acknowledged that unexpected risks could still emerge, particularly if the model is used alongside other tools.
“That uncertainty, along with the model’s broader step change in capabilities, is why we are pairing the model’s increased capabilities with stronger safeguards and a phased release,” the company said Friday.
