OpenAI Delays GPT-5.6 Public Release Following Federal Directive

Reports indicate that OpenAI has deferred the general public release of its next-generation large language model (LLM), GPT-5.6. This strategic delay follows a formal directive from the Trump administration, which seeks to restrict early access to a vetted cohort of government-approved entities. This move marks a pivotal moment in the tension between rapid technological iteration and the necessity of national security safeguards.

The decision highlights a growing trend regarding the intersection of frontier AI capabilities and systemic cybersecurity risks. According to a report by The Information on June 25, 2026, this intervention signals an era of increased federal oversight concerning the deployment of models that possess high-order reasoning and autonomous technical capabilities.

The primary driver for this scrutiny is the potential for advanced models to automate the entire offensive cyber kill chain. Specifically, there are significant concerns regarding the ability of such systems to perform automated vulnerability discovery, complex exploit development, and the orchestration of sophisticated, multi-stage cyberattacks. This shift suggests that regulatory bodies are moving toward a “pre-release oversight” posture to mitigate these high-impact risks.

This request follows recent regulatory friction between the White House and Anthropic. In that instance, the administration invoked export controls to temporarily suspend the operations of Anthropic’s Fable 5 and Mythos 5 models.

The Mythos 5 model, in particular, was distributed through a highly controlled deployment framework known as Project Glasswing to approximately 40 high-value organizations, ranging from Tier-1 financial institutions to major technology conglomerates. U.S. officials flagged Mythos 5’s ability to autonomously identify zero-day vulnerabilities and execute chained attack sequences—capabilities that represent a significant “dual-use” risk if they fall into the hands of adversarial nation-states.

Internal sources suggest that federal authorities view the architectural capabilities of OpenAI’s GPT-5.6 as being on par with the Mythos series. Consequently, the Office of the National Cyber Director (ONCD) and the Office of Science and Technology Policy (OSTP) have requested a staggered, permissioned rollout. Rather than a global API release, access is expected to be managed on a per-customer, vetted basis. Commerce Secretary Howard Lutnick has reportedly advocated for this inter-agency coordination to prevent a broad, unmonitored release of sensitive model weights or capabilities.

During an internal briefing, OpenAI CEO Sam Altman confirmed that GPT-5.6 will undergo a controlled preview phase, restricted to select enterprise partners. This phase will involve active participation from government agencies to vet recipients and ensure that the model’s deployment aligns with domestic security protocols.

While OpenAI is complying with these directives, the company has expressed reservations regarding the scalability of such a model. Altman has noted internally that while the current cooperative stance is necessary, a permanent, permissioned-access model may not be a sustainable framework for the long-term evolution and commercialization of AI technology.

The timeline for a broader public launch remains contingent on the results of this preview phase and subsequent federal approvals. A White House official stated that federal agencies are working alongside AI developers to establish a consensus on safety benchmarks as the ceiling for model autonomy continues to rise.

This situation underscores a critical regulatory vacuum in the United States. Currently, there is no standardized, legally mandated framework for the pre-release evaluation of frontier AI systems. While existing executive orders encourage voluntary cybersecurity reviews, they lack the enforcement mechanism required for mandatory compliance. The current arrangement with OpenAI—described as a cooperative, rather than legally binding, agreement—may serve as a prototype for future public-private coordination in managing high-stakes, dual-use technologies.