The United States government is quietly considering a move that could reshape the future of artificial intelligence: reviewing advanced AI models before they are released to the public. According to reports, the administration of Donald Trump is exploring an executive order to create a working group that would bring together government officials and leading AI companies to evaluate frontier systems ahead of launch.

At first glance, this looks like a technical policy shift. In reality, it marks something much bigger—a potential transition from a largely hands-off approach to one where governments actively monitor the most powerful AI systems before they reach the public domain. The proposal would reportedly involve companies such as OpenAI, Google, and Anthropic, with discussions centered around giving the government early access to models without necessarily blocking their release.

From “Let It Thrive” to “Let’s Take a Look First”

What makes this particularly notable is the reversal in tone. Since returning to office, Trump has been one of the strongest advocates for rapid AI development, previously warning against “foolish rules” that could slow the industry down. His administration rolled back earlier regulatory frameworks that required companies to share safety evaluations before releasing high-risk systems.

Now, that position appears to be evolving. The shift is being driven not by abstract concerns, but by the real-world capabilities of next-generation AI models. Reports suggest that a new system from Anthropic—referred to as Mythos—has raised alarms among cybersecurity experts due to its ability to identify vulnerabilities and potentially design sophisticated cyberattacks.

AI as a National Security Risk—and Asset

This is where the debate becomes more complex. AI is no longer just a commercial technology; it is increasingly viewed as a national security issue. Governments are beginning to recognize that the same systems capable of automating coding, research, and analysis could also be used to exploit software vulnerabilities, generate attack strategies, or accelerate cyber warfare.

The proposed review system reflects that dual reality. Officials are not only concerned about preventing AI-enabled attacks, but also about understanding how these capabilities could be leveraged by defense and intelligence agencies. In other words, the goal is not necessarily to restrict AI—but to see it first.

A New Model of Oversight Emerges

The idea of pre-release review is not entirely new. The United Kingdom has already been developing similar frameworks, where government bodies assess whether AI systems meet certain safety thresholds. The U.S. approach appears to be moving in a similar direction, though with a key difference: early indications suggest the government may not seek to block releases, but rather gain visibility into what is being deployed.

That distinction matters. It suggests a future where frontier AI models operate in a gray zone—technically private, but strategically visible to governments. This could create a new layer of coordination between tech companies and state actors, especially as the geopolitical competition around AI intensifies.

The Beginning of a New AI Era

The broader implication is clear: AI has reached a level of capability where governments can no longer afford to stay on the sidelines. The conversation is shifting from “should we regulate AI?” to “how do we manage systems that may outpace existing security frameworks?”

At the same time, there is a delicate balance to maintain. Too much oversight risks slowing innovation and pushing development offshore. Too little risks leaving governments unprepared for the consequences of increasingly autonomous systems.

The proposed working group may be an attempt to find that middle ground—one where AI continues to advance rapidly, but not entirely unchecked.

What happens next will likely define the relationship between governments and AI companies for the next decade. Because once states begin reviewing intelligence systems before release, the line between technology and infrastructure starts to blur.

Source: Reporting by The New York Times on White House AI policy discussions (2026).