OpenAI has begun a phased introduction of its latest GPT-5.6 AI model series, following consultations with the US government regarding the pre-release evaluation of advanced AI systems. This initial rollout will see the technology made available to a select group of trusted partners, with a broader distribution planned for the near future.
The decision to limit early access comes after discussions with government officials centered on AI safety, cybersecurity threats, and the responsible deployment of such technologies. OpenAI emphasized that this cautious approach is temporary, aiming to allow time for a thorough review process before more widespread access to the models is granted in the coming weeks.
The GPT-5.6 series comprises three distinct models: Sol, Terra, and Luna. Sol stands out as the most sophisticated of the trio, tailored for enhanced performance. In contrast, Terra and Luna are designed to offer more cost-effective solutions. OpenAI has stated that the most powerful model does not reach its internal criteria for posing significant cyber threats, instead being optimized for detecting and resolving security vulnerabilities rather than executing attacks.
While OpenAI supports the current limited release as a short-term measure, the company has voiced concerns about the potential for government approval processes to become a permanent fixture in AI model launches. OpenAI envisions this step as laying the groundwork for a more refined framework for future AI unveilings.
This development occurs amid heightened global scrutiny over advanced AI technologies, as governments grapple with issues related to cybersecurity, potential misuse, and the swift advancement of cutting-edge AI capabilities. The move reflects a broader trend of increasing regulatory attention aimed at managing the risks associated with next-generation AI systems.