OpenAI Delays GPT-5.6 Release Under White House Pressure

OpenAI confirmed Friday it is postponing the public launch of GPT-5.6, its next-generation AI model family, following a request from the Trump administration. The company will instead share the models initially with a limited set of customers pre-approved by the U.S. government before gradually expanding access.

The delay stems from an executive order President Trump signed earlier this month addressing cybersecurity concerns around powerful AI systems. That order established what it called a "voluntary process" for AI labs to share models with the government 30 days before public release. However, OpenAI executives said no formal voluntary framework actually exists yet, leaving frontier AI companies in an uncertain interim period where cooperation appears less than voluntary.

Why it matters

This marks a significant shift in how the U.S. government interacts with AI development. While the Trump administration has generally sought to reduce regulatory barriers to maintain competitiveness with China, these recent interventions create a de-facto approval process for model releases. The approach affects not just OpenAI but the entire U.S. AI industry—Anthropic was recently forced to take its most advanced models offline for all customers following a separate export control directive, with some of its own employees still unable to access the company's best systems.

The approval process

OpenAI plans to expand its approved customer list next week, including some international partners. Company executives said they submit customer lists to the government and receive feedback, though they cannot disclose details of how the White House evaluates these submissions.

In a blog post, OpenAI expressed clear dissatisfaction with 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."

The company characterized the delay as a short-term compromise to achieve broader availability in coming weeks while working with the administration to develop the cyber executive order framework and establish a repeatable process for future releases.

Three model variants

GPT-5.6 will launch in three versions: Sol, the most capable variant; Terra, a middle-tier option; and Luna, designed for speed and affordability. OpenAI says GPT-5.6 Sol represents its strongest performance yet on benchmarks testing cybersecurity, biology, and agentic capabilities. The company emphasized it has implemented a "layered safeguard stack" intended to prevent malicious use, including cyberattacks.

The situation creates uncertainty for other U.S. AI labs navigating the gap between the administration's stated goal of clearing regulatory obstacles and its growing concern about AI-enabled cybersecurity threats.

These details were first reported by The Information and confirmed by WIRED.