OpenAI said it will begin with a limited preview of its GPT-5.6 model family, making the technology available only to a small group of trusted partners whose participation has been shared with the US government. The company said it expects broader availability in the coming weeks as it works with the administration on a framework for future releases.
The move follows a request from the White House as federal agencies develop a process to evaluate the cybersecurity and national security implications of increasingly powerful AI models before they are widely deployed. It represents one of the first instances of the US government seeking to influence the public rollout of a frontier AI model before its general release.
According to reports, the request came after discussions involving the White House Office of the National Cyber Director, the Office of Science and Technology Policy, and other government agencies responsible for AI and cybersecurity. The review is intended to help establish a repeatable process for testing advanced AI systems before they become broadly accessible.
OpenAI stated that it does not view this approach as the preferred long-term model for releasing AI systems. In a company statement, it said it believes keeping advanced AI tools away from developers, enterprises and cybersecurity defenders should not become the default approach. However, it described the limited rollout as a short-term measure while collaborating with the US administration on a broader security framework.
The development comes weeks after President Donald Trump signed an executive order directing federal agencies to create a voluntary framework for evaluating advanced AI models before public deployment. The order aims to strengthen cybersecurity while maintaining US leadership in artificial intelligence, balancing innovation with national security considerations.
The phased rollout also follows heightened government attention on frontier AI models after similar concerns emerged around Anthropic's latest systems. The broader policy shift reflects growing worries among policymakers that increasingly capable AI models could be exploited for sophisticated cyberattacks, vulnerability discovery, or other security risks if released without adequate safeguards.
OpenAI's latest release includes multiple variants designed for different use cases, ranging from complex reasoning tasks to faster, lower-cost deployments for enterprise workloads. While only selected organisations will receive early access, the company indicated that wider commercial availability remains the objective once government review processes are completed.
The decision highlights the evolving relationship between AI developers and governments as frontier models become more capable. Until recently, major AI companies largely determined their own release schedules, relying on internal safety evaluations and external testing. The latest development suggests governments may play a more active role in overseeing the deployment of highly capable AI systems.
For the AI industry, the move could set an important precedent. Other frontier AI developers, including Google, Anthropic and xAI, may face similar expectations as governments around the world seek stronger oversight of advanced AI technologies without introducing formal licensing regimes.
The episode also underscores the growing challenge of balancing rapid AI innovation with security concerns. As competition intensifies among US and global AI companies, policymakers are increasingly exploring ways to ensure that the deployment of powerful AI models does not outpace the safeguards designed to manage their potential risks.