Trump Administration Restricts OpenAI's GPT-5.6 Release to Government Partners

The Trump Administration has imposed a restriction on OpenAI’s upcoming GPT-5.6 model, limiting its release to a select group of government-approved partners. This decision, announced via a White House memo, stems from concerns that the model’s advanced cybersecurity capabilities could pose unprecedented risks. OpenAI agreed to the condition as a pathway to eventual public availability, but the move underscores a critical flaw: the U.S. lacks a transparent framework for overseeing AI model safety.

According to a memo shared by OpenAI CEO Sam Altman, the government now approves each company’s access to advanced models ‘customer by customer.’ This approach, while intended to mitigate risks, creates an opaque process where private entities like OpenAI must navigate bureaucratic hurdles without clear guidelines. The White House emphasized that GPT-5.6 shares similarities with Anthropic’s Mythos model, which was withdrawn in June due to cybersecurity concerns. This comparison suggests regulators view GPT-5.6 as similarly high-risk, though OpenAI disputes the equivalence.

OpenAI’s compliance comes amid broader pressure on AI labs to adhere to an executive order requiring 30-day reviews of advanced models before public release. However, the mechanism for this review remains unimplemented, leaving the process ad hoc. Altman expressed frustration in a memo, stating the administration’s approach is ‘not our preferred long-term model’ and urging collaboration to establish sustainable oversight. This tension highlights a clash between rapid innovation and regulatory caution.

Experts argue the restriction reflects a reactive strategy rather than proactive governance. Brad Carson of Public First, a pro-AI safety group, criticized the lack of transparency, calling the current system ‘ad hoc, personalised, opaque, possibly lawless.’ Without standardized protocols, such restrictions could set a precedent for arbitrary control over AI development. Competitors like Google and Anthropic may gain an edge if OpenAI’s delays hinder its market position.

The executive order itself is a response to growing fears about AI’s potential misuse. However, its effectiveness is undermined by the absence of clear criteria for evaluating risks. Cybersecurity experts note that while the goal of preventing harm is valid, the current process lacks accountability. For instance, there’s no independent body to verify claims about a model’s capabilities or enforce compliance.

This situation also raises questions about the future of AI regulation. If the U.S. continues to rely on voluntary compliance and fragmented reviews, other nations with stricter frameworks—such as the EU—could dominate AI governance. OpenAI’s experience may accelerate demands for a global regulatory body, though such a move faces political and technical challenges.

The immediate impact on OpenAI remains uncertain. While the company has not commented beyond Altman’s memo, delays could affect its competitive stance. Meanwhile, the restriction sets a precedent for how governments might manage emerging technologies. Critics warn that without clear rules, such actions could stifle innovation or be weaponized for political purposes.

Ultimately, the GPT-5.6 restriction exemplifies the broader challenge of governing AI. As models grow more powerful, the line between safety and control becomes blurred. The U.S. must decide whether to adopt a precautionary approach or risk falling behind in both regulation and technological leadership.