What's Actually Happening

President Trump is expected to sign an executive order on AI and cybersecurity as soon as this week. According to Reuters and the Daily Signal, the order will create a voluntary framework asking AI developers to submit their most powerful models to the government 90 days before public release.

The Office of the National Cyber Director gets 60 days to develop the actual framework. Banks and other critical infrastructure providers would also get pre-release access under the plan.

The draft does not address what happens if the government reviews a model and finds it dangerous, sources familiar with the order told the Daily Signal. The framework creates the vetting process but omits enforcement mechanisms for a failed review.

What Forced This

Anthropic released Claude Mythos in April 2026. According to multiple reports, the model exposed thousands of vulnerabilities in major software and critical digital infrastructure. That grabbed Washington's attention immediately.

OpenAI's GPT-5.5-Cyber is also in the mix. Both companies have warned these new models could supercharge complex cyberattacks. Some cybersecurity executives say those fears are overblown — but nobody's putting their name on that reassurance publicly.

The result: a genuine national security debate broke out inside Trump's own coalition.

A Fight Inside MAGA World

This isn't Republicans vs. Democrats. It's a civil war among Trump's own supporters.

On one side: Steve Bannon and political organizer Amy Kremer, pressing the White House for mandatory government security testing before any powerful AI model ships. They want real teeth in the rules.

On the other side: venture capitalist Marc Andreessen and former Trump AI czar David Sacks, who oppose mandatory requirements. Sacks stepped down from his White House AI role in March 2026 and now co-chairs Trump's tech advisory committee — meaning he's still in the room.

Trump's second-term AI policy has, until now, heavily reflected the Andreessen-Sacks camp. He repealed Biden's AI safety executive order on day one. His own order removed "barriers" to AI development. He signed a separate order threatening states trying to maintain their own AI regulations.

The voluntary framework is a split-the-difference move. It gives the safety-hawk populists something they can point to. It gives the tech-friendly libertarians nothing mandatory to complain about.

What the Polling Actually Shows

The Institute for Family Studies, working with YouGov, found that 82% of Americans support mandatory AI safety testing. Trump voters polled even higher — 90% in favor.

Eighty-eight percent of Americans want AI systems tested specifically for national security risks. Eighty-seven percent favor safety testing for risks to children and families.

A separate 2025 Gallup poll found 80% of U.S. adults prioritize data security and AI safety rules over fast-paced innovation, according to Built In.

The public wants mandatory testing. The order delivers voluntary testing.

What's Already in Place — and What Got Quietly Deleted

The Center for AI Standards and Innovation — CAISI — already exists inside the Commerce Department. It's the Trump administration's renamed version of Biden's U.S. AI Safety Institute. According to Fortune, CAISI has completed more than 40 pre-deployment evaluations, including on unreleased state-of-the-art models.

On May 5, Google DeepMind, Microsoft, and xAI agreed to work with CAISI. Then the Commerce Department deleted the details of that arrangement from its webpage, according to the Daily Signal. No explanation was provided.

White House National Economic Council Director Kevin Hassett told Fox Business the administration is considering an order to create "a clear road map" — comparing the process to FDA drug approval. The FDA does not simply test drugs and move on when they fail.

What the Media Is Getting Wrong

Left-leaning coverage frames this entirely as Trump finally being dragged toward the Biden position on AI safety. Fortune's headline captures it: the administration is "implementing many of the same policies" it once rejected.

That framing isn't wrong, but it's incomplete. The Biden AI safety executive order had its own problems — vague mandates, sprawling scope, and little enforcement mechanism. The fact that Trump is landing near a similar spot doesn't mean the destination is correct. Both administrations are dancing around the same hard question: what do you actually do when an AI model is dangerous?

Right-leaning coverage focuses on the populist vs. tech-libertarian tension inside the GOP, which is the more honest framing of where the real fight is happening.

Neither side is spending much time on the deleted Commerce Department webpage. The circumstances surrounding that deletion warrant scrutiny.

What This Means for You

If you work in finance, energy, healthcare, or any other critical infrastructure sector, a powerful AI model that maps your vulnerabilities could go public with 90 days' notice — and if the government review finds a problem, there's currently no legal mechanism to stop the release.

If you're a parent, the order doesn't address AI companions pushing kids toward self-harm, AI tools helping plan violence, or the documented mental health damage from chatbot dependency. The Institute for Family Studies polling flagged all of this. The executive order doesn't touch it.

If you're a taxpayer, the government is hiring more people to help with cyber preparedness under this order. How many, at what cost, with what accountability — not specified yet.

A voluntary framework with a 60-day runway to write the actual rules is better than nothing. It falls short of the mandate 82% of Americans indicated they wanted.