OpenAI Has New AI Models. Here’s Why You Can't Use Them

OpenAI is delaying the nationalist merchandise of its adjacent procreation of AI models, GPT-5.6, astatine the petition of Trump’s White House, the institution confirmed connected Friday. OpenAI said it would archetypal stock the models with a tiny acceptable of customers, which volition beryllium pre-approved by the US government. It volition past enactment with the medication to dilatory grow access.

OpenAI is not blessed astir this, according to a idiosyncratic acquainted with the company’s thinking, but believes the hold and authorities support process is lone temporary. In a blog post, the institution said it hopes it volition beryllium capable to marque GPT-5.6 disposable to everyone successful the coming weeks. OpenAI’s plans to hold its adjacent procreation of AI models astatine the Trump administration’s petition was archetypal reported by The Information.

“We don’t judge this benignant of authorities entree process should go the semipermanent default,” OpenAI wrote successful its blog post. “It keeps the champion tools from users, developers, enterprises, cyber defenders, and planetary partners who request them. We are taking this short-term measurement due to the fact that we judge it is the strongest way to broader availability successful the coming weeks, portion we enactment with the Administration to make the cyber Executive Order model and a repeatable process for aboriginal exemplary releases.”

Earlier this month, President Trump signed an enforcement bid that aimed to code the cybersecurity concerns of almighty caller AI models. The bid said the White House would make a “voluntary process” for AI labs to stock their models with the authorities 30 days up of a broader release. The mandate included a carveout, saying the US authorities would not crook its voluntary process into a de-facto “licensing regime.” But successful its Friday briefing, OpenAI executives said nary specified voluntary model exists yet. As a result, the frontier AI labs are successful a precise weird interim period, wherever moving with the US authorities connected your AI exemplary motorboat doesn’t look each that voluntary.

The White House is asking OpenAI to stagger the merchandise of its AI models conscionable 2 weeks aft it sent an export power directive to Anthropic, which prompted the institution to instrumentality its astir precocious AI models offline for each customers. Anthropic’s spat with the White House is inactive unresolved, and immoderate of the company’s ain employees are inactive barred from utilizing its astir precocious AI models.

The Trump Administration’s petition for OpenAI and Anthropic to bounds availability connected their astir precocious AI models creates an uncertain situation for different US AI labs. Over the past 2 years, the Trump medication has sought to wide regularisation and reddish portion that could hinder America’s AI innovation, and perchance wounded the country’s competitiveness with China. In caller months, however, the White House has grown progressively acrophobic astir the cybersecurity abilities of caller AI models, and has scrambled to code the problem.

OpenAI plans to broaden the acceptable of customers it tin stock GPT-5.6 with adjacent week, including immoderate planetary partners. Executives for OpenAI said that it can’t stock details of however precisely the White House is approving these customers—the institution conscionable sends the US authorities a list, and past gets feedback connected it, the executives said.

The White House didn’t instantly respond to WIRED’s petition for comment.

OpenAI says its GPT-5.6 AI models volition travel successful 3 flavors: Sol, its astir susceptible mentation of the model; Terra, a middle-tier mentation of the model; and Luna, a accelerated and affordable version. The institution says GPT-5.6 Sol is its astir susceptible exemplary yet connected benchmarks investigating cybersecurity, biology, and agentic abilities. Alongside these caller capabilities, OpenAI says it has a “layered safeguard stack,” which aims to halt atrocious actors from utilizing its AI exemplary for cyberattacks, among different malicious behaviors.