OpenAI limits new AI models to 'trusted partners' at request of U.S. government

On Friday, OpenAI unveiled three new artificial intelligence models, announcing that their initial release will be restricted to a select group of 'trusted partners' following a request from the U.S. government. In a detailed blog post, OpenAI expressed its commitment to providing broad access and confirmed plans to make models—GPT-5.6 Sol, Terra, and Luna—available to the public in the coming weeks. The company emphasized that it had shared the capabilities of these models with government officials prior to the launch. OpenAI stated, 'We don't believe this kind of government access process should become the long-term default,' arguing that such limitations hinder essential tools from reaching users, developers, enterprises, cybersecurity professionals, and international partners who require them. Although OpenAI has not revealed which partners will have access to the new models, the announcement arrives shortly after competitor Anthropic was forced to disable access to two of its latest offerings to comply with an export control directive issued by the previous administration. Anthropic is currently in discussions with Washington officials and has not provided a timeline for when its models will be operational again. The Trump administration's proactive stance on AI regulation has become evident, particularly after the issuance of an executive order that encourages AI developers to allow government assessments of their models before full-scale releases. OpenAI is collaborating with the administration to establish a regulatory framework and create a consistent process for future model launches. In their announcement, OpenAI described the GPT-5.6 Sol model as their most powerful yet, with advancements in coding and biology, and highlighted its superior capacity for cybersecurity tasks. The company reassured users that while the model excels at identifying and fixing vulnerabilities, it does not engage in severe attacks and remains below the threshold for what it defines as 'critical' cybersecurity risks, which could lead to significant harm.