OpenAI Limits GPT-5.6 Rollout After U.S. Government Request

OpenAI has started the GPT-5.6 rollout as a limited preview after a request from the U.S. government, slowing wider access to its newest model family while officials review security concerns. The company said OpenAI is first making GPT-5.6 available to a small group of trusted partners whose participation has been shared with the government.

The decision affects GPT-5.6 Sol, Terra, and Luna, not a single model called “ChatGPT 5.6.” OpenAI says broader access for ChatGPT, Codex, and the API is still planned soon, but the company has not opened the models to all users yet.

Reuters reported that the full public launch was deferred at the government’s request, with initial access limited to vetted partners. The move shows how frontier AI releases are becoming a more direct national security issue in Washington.

What happened with GPT-5.6?

In its launch post, the company said the GPT-5.6 series includes Sol as the flagship model, Terra as a balanced everyday model, and Luna as a faster, lower-cost option. OpenAI said Terra is designed to compete with GPT-5.5 at a lower price, while Luna targets speed and affordability.

The limited preview is available through the API and Codex for selected partners and organizations. OpenAI says it plans broader availability for people using ChatGPT, Codex, and the API soon, but the staged rollout gives the company and U.S. officials more time to evaluate safeguards.

The Reuters report said U.S. officials want early access to frontier models to identify possible threats before wide deployment. Those concerns include cyber misuse, military misuse, and other risks tied to more autonomous AI systems.

Why did the U.S. government ask for a slower rollout?

The request comes as the Trump administration builds a more formal process for reviewing advanced AI systems before wider release. Axios reported that the White House Office of the National Cyber Director and the Office of Science and Technology Policy asked OpenAI to limit the rollout while the administration develops a testing framework.

The White House has already moved in this direction through an executive order on advanced AI innovation and security. That order says advanced AI capabilities strengthen the country, but also introduce national security considerations that require coordinated action across federal agencies.

The White House order does not create a full public licensing system for AI model launches. Instead, it points to cooperation between government and industry, especially around cyber defense, critical infrastructure, and protection of U.S. technology from foreign misuse.

How Anthropic’s Mythos case shaped the OpenAI decision

The OpenAI decision follows a separate government action involving Anthropic. In a June 12 statement, Anthropic said the U.S. government issued an export control directive requiring the suspension of access to Fable 5 and Mythos 5 by foreign nationals, including foreign national employees.

Anthropic said the practical effect was that it had to disable Fable 5 and Mythos 5 for all customers to ensure compliance. According to Anthropic’s statement, the directive did not affect access to its other models.

The comparison matters because Mythos 5 was positioned as an advanced model for selected cyberdefense and infrastructure users. OpenAI’s GPT-5.6 Sol is also described as a major step forward in cybersecurity, coding, scientific work, and long-horizon agentic tasks.

What GPT-5.6 can do

OpenAI’s GPT-5.6 Preview System Card describes the family as a meaningful step up in cybersecurity capability. The company says the models can help find vulnerabilities and pieces of exploits, but did not reach the highest “Critical” risk level in its framework under the tested conditions.

The system card also says GPT-5.6 Sol and Terra could not carry out autonomous, end-to-end attacks against hardened targets during the company’s cybersecurity testing. OpenAI still warns that benchmarks cannot represent every real-world workflow, tool combination, or misuse scenario.

Model	Positioning	Main focus
GPT-5.6 Sol	Flagship model	Advanced reasoning, coding, science, cybersecurity, and agentic work
GPT-5.6 Terra	Balanced model	Everyday professional work at lower cost than GPT-5.5
GPT-5.6 Luna	Fast, lower-cost model	High-volume tasks where speed and price matter

Who can use GPT-5.6 now?

For now, access remains limited to selected trusted partners and organizations. OpenAI has not publicly named the full list of preview customers, and the broader release timeline depends on the preview period and continued coordination with the government.

• GPT-5.6 is currently in limited preview, not full general release.
• Initial access is available through API and Codex for selected partners.
• Broader access for ChatGPT, Codex, and API users is planned soon.
• The U.S. government is reviewing frontier model risk as part of a developing AI security process.
• Axios reported that OpenAI had been working with the administration on the release before the Anthropic incident became public.

The key takeaway is clear: OpenAI has not canceled GPT-5.6, but it has limited the first phase of access. The company wants to move toward broader availability while also avoiding a permanent system where government agencies approve customers one by one.

Why this matters for AI users and developers

The GPT-5.6 rollout could set an important precedent for future AI releases in the United States. If the limited preview becomes a common model, major labs may need to build more time into launches for government review, partner vetting, and security testing.

For developers, enterprises, and cybersecurity teams, the delay could mean slower access to more powerful tools. For policymakers, it offers a chance to test whether voluntary review can reduce risk without turning frontier AI deployment into a closed approval system.

OpenAI’s position is that short-term coordination may help speed broader availability, but that government access reviews should not become the default way new AI models reach the market. The system card makes the same broader point in technical terms: stronger capabilities require stronger safeguards, but defenders also need access to useful tools.

FAQ