APREX · EST. 2026
Recent weeks have made it clear that the AI race in the United States is now unfolding under greater political scrutiny than ever before. According to Wired, the White House asked OpenAI to hold back its new GPT-5.6 models — just two weeks after Anthropic was forced to take its most advanced models offline under similar circumstances.
An Executive Order Sets the Framework
The backdrop is an executive order signed by Donald Trump on June 2, 2026, titled "Promoting Advanced Artificial Intelligence Innovation and Security." The order establishes a framework under which AI developers may voluntarily submit their frontier models for government review for up to 30 days before public release.
The aim is to map advanced cybersecurity capabilities and identify potential security vulnerabilities. The Departments of Treasury and Defense, along with the National Security Agency (NSA), have been tasked with developing a classified evaluation process to determine which models fall under the category of "covered frontier model."
Anthropic's Mythos Model Triggered the Concerns
According to source material, it was the limited release of Anthropic's Mythos model that fundamentally shifted the administration's approach. The model reportedly demonstrated a remarkable ability to identify and exploit serious software vulnerabilities. OpenAI's own GPT-5.5-Cyber also contributed to heightening those concerns.
It is against this backdrop that the White House is now asking OpenAI to hold off on GPT-5.6.
"The US should lead on AI by continuing to develop the best models, ensuring they are safe, and providing cyber tools to trusted defenders" — Sam Altman, OpenAI
OpenAI and Experts Split on the Assessment
OpenAI CEO Sam Altman has, according to source material, expressed support for the executive order, saying it "strikes the right balance." Not everyone shares that view, however.
Daniel Castro of the think tank Information Technology and Innovation Foundation (ITIF) warns that the order could in practice require companies to seek government permission to innovate. He argues this would shift the pace of innovation from Silicon Valley to Washington — with a potentially chilling effect.
On the other side, Chris McGuire of the Council on Foreign Relations sees the measure as a long-overdue policy shift. He stresses that a clear plan for security after models are released is also needed, including requirements for cyber and physical security at AI laboratories and cloud providers.
White House spokesperson Liz Huston describes the order as "a sensible approach that balances innovation and security, and ensures America's continued global dominance in AI and cybersecurity."
Voluntary — But With Political Weight
It is worth noting that the executive order explicitly states it should not be interpreted as authority for mandatory licensing or pre-approval of AI models. Formally speaking, this is a voluntary regime.
Nevertheless, the cases of OpenAI and Anthropic demonstrate that the White House is willing to apply political pressure to bring companies into compliance with the framework — even without legally mandated sanctions.
Federal agencies are required to begin preparations within 30 days of the order being signed, meaning by July 2, 2026. The benchmarking process and voluntary framework itself carry a 60-day deadline, with deliverables expected by August 1, 2026.
When GPT-5.6 will actually become available to the public has not been confirmed as of the publication date.