OpenAI Restricts GPT-5.6 Release to Government-Approved Partners

OpenAI said Friday it is restricting access to GPT-5.6, its newest and most capable family of artificial-intelligence models, to a small group of partners approved by the Trump administration. The company is releasing three versions through its API and the Codex coding tool, the flagship Sol, a midtier Terra and a low-cost Luna, but not through ChatGPT during the preview. Axios reported that the arrangement marks the first time the U.S. government has preemptively asked an American AI company to limit a model's release before launch.

The preview reaches roughly 20 organizations, according to Axios and the AP. OpenAI said the partners' participation has been shared with the government and that it plans to broaden access, including to some international partners, next week. In a memo to staff first reported by The Information, Chief Executive Sam Altman said the government would be "approving access customer by customer during this preview period" and that he hoped for a wider release "a couple of weeks later." Altman discussed the rollout with Commerce Secretary Howard Lutnick on Wednesday, and the request came after conversations with the White House's Office of the National Cyber Director and Office of Science and Technology Policy, according to Axios.

In the same blog post, OpenAI objected to the arrangement. "We don't believe this kind of government access process should become the long-term default," the company wrote. "It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them." It described the restriction as a short-term step while it works with the administration on a cybersecurity framework required under an executive order President Trump signed earlier this month, which gives the government up to 30 days to review frontier models before release.

OpenAI positioned Sol as its strongest model yet, with gains in coding, biology and cybersecurity. On Terminal-Bench 2.1, a test of command-line workflows, the company reported Sol scoring 88.8% and a new subagent "ultra" mode reaching 91.9%, ahead of Anthropic's Claude Mythos 5 at 88.0%. GPT-5.6 also introduces a "max" setting for longer reasoning. Priced per million tokens, Sol costs $5 for input and $30 for output, Terra runs at half that, and Luna at $1 and $6.

A source told Axios the administration intervened because GPT-5.6 has "Mythos-like" capability. OpenAI said Sol matched the performance of an earlier Anthropic model, Mythos Preview, on the ExploitBench security benchmark while using about a third of the output tokens. In tests against the Chromium and Firefox browsers, OpenAI said Sol identified software bugs and the building blocks of an exploit but did not autonomously produce a working full-chain attack, leaving it below the "Cyber Critical" threshold in the company's internal preparedness framework. OpenAI said it dedicated more than 700,000 A100-equivalent GPU hours to automated red-teaming aimed at finding jailbreaks that work across many prompts, and that Sol is "better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks."

The approach drew criticism. Dean Ball, a former White House AI adviser who is joining OpenAI, said Trump's executive order had created a de facto involuntary licensing regime for frontier AI. Alex Stamos, the chief product officer at security company Corridor and a former chief security officer at Meta, told reporters this week that "pretty much nobody in the cybersecurity industry believes that there's any factual basis for this action," and said gating models this way would set back the United States in its competition with China. Representative Lori Trahan, a Massachusetts Democrat, said in a statement that the administration was deciding "company by company who gets access to the newest AI model," with "no law, no process, no oversight."

The restriction follows a similar standoff over Anthropic's models. A Commerce Department directive two weeks ago forced Anthropic to pull its Fable 5 and Mythos 5 systems over their cyber capabilities. The government lifted part of that restriction Friday, allowing Mythos 5 to be redeployed to a limited group of cyber defenders and infrastructure providers, Anthropic said. The department's Center for AI Standards and Innovation has been reviewing GPT-5.6 as it tests other models, people familiar with the matter told The Wall Street Journal.

OpenAI said it plans to make GPT-5.6 generally available through ChatGPT, Codex and the API in the coming weeks, and to bring Sol to Cerebras hardware at up to 750 tokens per second in July.

NSA Loses Anthropic Mythos Access After June Export-Control Order

The National Security Agency lost access to Anthropic's Mythos 5 after the Trump administration's June export-control directive forced the company to pull back its most advanced models, The New York T

The Implicator

The White House Is Asking Anthropic for the Impossible

A government can stop a risky AI model from reaching foreign users. But perfect jailbreak resistance is not a compliance standard any lab can reliably meet, and that is the direction officials now app

The Implicator

The Anthropic Shutdown Confirmed Europe's Fear of Depending on US Tech

Anthropic disabled its two most capable artificial-intelligence models, Fable 5 and Mythos 5, for every customer worldwide late on June 12, after the US Commerce Department ordered the company to bloc

The Implicator

Article Components

Image Prompt

A glowing trio of celestial orbs (a bright sun, a blue earth, and a small moon) representing three AI models, lined up behind a government checkpoint barrier with an official eagle seal, while a uniformed official at a booth checks a clipboard and a small line of business figures in suits waits to be approved through the gate.

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai