Trump writes new rules to stop rogue AI — but they’re classified

President Trump’s approach to AI regulation has been mixed. First, he limited states from making their own laws in favor of a national framework. Then, despite proposed laws to protect consumers, no federal bills have passed yet.

Now, as advanced AI threatens the cybersecurity of government institutions and private corporations, Trump is moving forward with a government-sanctioned review process for frontier AI models, and employees at the major AI giants are all begging for it.

Trump chose not to share the details of his AI review framework with the public.

Here’s how the White House finalized its AI review framework.

Behind the curtain

The government review process was originally announced in June. At the time, each major AI company could willingly allow the government to review its latest AI models to check for potential cybersecurity risks before they were released broadly to the public. Fast-forward a couple of months later, and now a more official structure is in place.

For whatever reason, President Trump chose not to share the details of his AI review framework with the public.

Leakers will leak, however, so we’re not left completely in the dark. Axios spoke with three sources connected to the executive order and uncovered several key points.

• Targeted AI models: The framework is aimed at reviewing closed-source frontier models with advanced capabilities that could pose a risk to national security. Open-source models are exempt from review.
• Limited access: New AI models set forth for review will be hosted in secure environments. While multiple administration officials will have access clearance, AI company employees are limited from using the new models for 30 days during the review cycle.
• Classified details: The process that the Trump administration will use to test AI models, as well as the AI review framework itself, will remain classified and kept from public knowledge.

As it stands, the review process is still voluntary. However, the exemption of open-source models makes it sound as if compliance for closed-source frontier models is strongly encouraged.

Arbitrary guardrails spell trouble for American AI dominance

Trump’s AI review framework comes at a time when employees at some of the largest AI companies in the U.S. have banded together to demand government regulation. In an open letter, workers from OpenAI, Anthropic, Google, and Meta urged the government to “deliberately pace” the development and progression of AI due to the recent security implications spurred on by Anthropic’s Mythos and other emerging threats.

Of course, slowing the pace of AI development is a double-edged sword.

RELATED: Rand Paul exposes the government’s cozy covid links to Bill Gates

David Paul Morris/Bloomberg/Getty Images

On one hand, arbitrary pauses, like the 30-day review period enforced by Trump’s AI review framework, would give institutions more time to protect themselves against potential vulnerabilities from new AI models. With a heads-up from the government, they can learn about any problems identified during the review and shore up security to ensure bad actors who get ahold of the latest models can’t use them to hack and exploit our most precious agencies, institutions, and establishments.

A tough spot

On the other hand, the pause window gives competing nations like China more time to steal from American tech companies and potentially overtake America as the leader in the AI race. At this point, our nation would become a prime target for foreign cyberattacks capable of breaching our systems before we even know what happened, leaving us with inferior intelligence and little-to-no means to stop it.

The only way to combat threats posed by Chinese AI is to speed up development instead of slowing it down, ensuring that American technology is always ahead of the nations that would rather see us fall.

The choice isn’t easy, as there are stark pros and cons to both sides. However, President Trump clearly believes that cyber threats posed by America’s homegrown AI models are a bigger problem for the nation than China overtaking the United States in AI development. There may be some merit to this decision, too. Considering that most of China’s AI innovations came from copying American AI companies, the former may not be as advanced as they want us to believe. If American AI companies suddenly stopped progressing as quickly, China’s development would likely slow down, as well, in which case, 30-day pauses to review new frontier models are undoubtedly the right call.