OpenAI began a limited preview release of its latest AI model, GPT-5.6, only to a small group of trusted partners after a request from the US government. Access during the preview is being approved on a customer-by-customer basis, with a broader release planned if the process proceeds smoothly, CEO Sam Altman said [1, 2, 3].
This staggered rollout followed talks between OpenAI and government bodies including the Office of the National Cyber Director and the Office of Science and Technology Policy [1, 2]. Altman acknowledged the approach was not OpenAI's preferred long-term plan but said the company would cooperate with regulators and industry to find a sustainable path for model releases. "We've made clear to the US government that this is not our preferred long term model..." he said [1].
The timing follows an executive order President Donald Trump signed on June 22 that requires AI companies to submit advanced models voluntarily for government review 30 days before public release [1, 4, 5]. The administration has shifted in 2026 from deregulation toward pushing for federal oversight and safety vetting of AI systems [1, 2].
Anthropic, a competitor, previously faced government orders to restrict access to its Mythos model amid concerns it included cyber-hacking capabilities such as discovering and exploiting software vulnerabilities rapidly [1, 2, 3]. The company has since removed Mythos from public access, as directed by regulators [1, 4]. OpenAI and government sources agree GPT-5.6's capabilities are roughly comparable to Anthropic's model [4, 3, 5].
Anthropic CEO Dario Amodei publicly called for government powers to block deployment of risky new AI, saying, "If a new generation of AI models causes specific risks, the government should have the power to stop AI developers from deploying those models" [3]. Whether OpenAI’s leadership discussed this publicly has been reported differently across sources [6, 3].
A White House official said the administration would continue collaborating with leading AI labs to develop unified approaches to the challenges posed by rapidly advancing AI technologies [4]. Meta is also reportedly under pressure to submit its Muse Spark AI model for security review [7].
Altman said the US government will review and approve GPT-5.6 users during the preview, with a broader rollout expected "a couple of weeks later" if no issues arise [1]. The preview release began on June 24–25, shortly after the executive order was signed.
OpenAI faces a regulatory environment that is shifting toward voluntary federal oversight with an emphasis on safety and gradual deployment of powerful AI models. The company and the government are expected to continue working together to establish a sustainable process for releasing future AI models safely.
