President Trump signed an executive order Tuesday asking AI companies to voluntarily submit their most powerful new models for federal testing up to 30 days before release. The order also directs the Treasury, the NSA, CISA, and NIST to develop AI cyber benchmarks and creates an "AI cybersecurity clearinghouse" to share vulnerability information. Crucially, it explicitly states that nothing in the order authorizes the creation of a "mandatory governmental licensing, preclearance, or permitting requirement" for new AI models. Trump scrapped an earlier draft less than two weeks ago that would have required a 90-day review. Trump's AI czar and venture capitalist David Sacks is widely credited with negotiating the window down to 30 days and the framework to voluntary.

1. This Is The Right Balance Between Innovation and Safety (Sacks, Andreessen, White House)

The most powerful models get a 30-day look. A licensing regime would have killed shipping. This is well calibrated.

Mandatory rules would entrench incumbents and push frontier work offshore. A voluntary 30-day window gives the government visibility into the largest models without creating a permitting bottleneck that only the biggest labs could afford to navigate.

The no-mandate clause is the point. We can't set a precedent for, say, a future administration quietly converting the framework into a permitting system. This may be the order's most important provision. The White House framed it as "common-sense" balance between innovation and security.

2. Safety Theater (critics, AI safety advocates)

Voluntary means optional. The window is too short. And the no-mandate clause kills enforcement before it can start.

The numbers are lame. The review window was cut from 90 days to 30; the framework dropped any compliance mechanism; the explicit no-mandate language forbids enforcement. Engadget called the result "scaled-back"; Scientific American said the order "drastically shifts" the administration's stance — toward weaker oversight.

But hey, the cybersecurity stuff is cool. Federal agency cyber-defense improvements are concrete. This may be safety theater, but at least there's one real thing in it.

3. A Quiet Big AI Win (Sacha Haworth, Tech Oversight Project)

A VC wrote the rules for an industry he's invested in. The same week Anthropic filed for a $965B IPO.

This is industry capture, plain and simple. Sacks himself has portfolio exposure to AI companies, and the explicit "no mandatory licensing" language locks in the AI industry's preferred regulatory posture for the duration of this administration.

Sacha Haworth made the case when Trump scrapped the May draft. Haworth told the press at the time that the White House "keeps handing Dems opportunities on silver platters to point out how Big Tech CEOs being in bed with lawmakers means the American people lose" — and the new voluntary order extends exactly that pattern, with Anthropic's confidential $965 billion IPO filing landing the day before.

Where This Lands

Trump signed a voluntary 30-day AI review on Tuesday and explicitly disclaimed any future government licensing. Sacks and the White House call it the smart balance between innovation and safety; critics call it the appearance of safety with most of the substance stripped out; and the structural reading is that Sacks — a VC with AI exposure — negotiated the terms the industry wanted.

Sources