OpenAI has initiated a selective rollout of its new GPT-5.6 AI model series following discussions with the US government. This move is part of a collaborative effort to scrutinize powerful AI systems before they are made publicly available. Initially, access to the new models will be granted to a limited group of trusted partners, with plans to broaden availability in the near future.
The decision to adopt this temporary measure comes after government talks that highlighted concerns around AI safety, cybersecurity threats, and the responsible deployment of advanced technologies. OpenAI indicated that a more widespread release of the GPT-5.6 series is anticipated in the coming weeks, contingent on the completion of a thorough review process.
Included in the GPT-5.6 series are three distinct versions: Sol, Terra, and Luna. The Sol model is the most advanced, offering superior performance, while Terra and Luna are designed to provide more cost-effective access. OpenAI has stated that the capabilities of the most robust model do not meet the company’s internal criteria for posing highly dangerous cyber threats. Instead, it is optimized for identifying and mitigating security vulnerabilities, rather than executing attacks.
While OpenAI acknowledges the necessity of government evaluations in the short term, it also expressed concern that such approval processes should not be a permanent fixture in the release of AI models. The company views this limited launch as a temporary step, aimed at developing a more effective framework for the future introduction of AI technologies.
This development comes amid increased governmental scrutiny over advanced AI technologies, driven by growing concerns about cybersecurity, potential misuse, and the rapid advancement of next-generation AI systems. Governments are keen to ensure that the deployment of such technologies does not compromise security or lead to unintended consequences.
