Zuckerberg, Musk, and Huang beat back an AI regulator

19 Sep 2026, 20:37 · The Fresno Bee

Markets almost never get a referee before they need one.

The Securities and Exchange Commission showed up after 1929 had already wiped out a generation of savers. The private body that polices your broker today grew out of that same wreckage.

That order of operations is expensive. By the time the rules arrive, the losses have been spread across everyone, and the people who absorbed them rarely got a vote.

Artificial intelligence (AI) has never had a referee at all. No federal agency tests a frontier model before it ships. No independent group verifies that a system can do what its maker claims, or something its maker never intended.

The only check that exists is a company’s own word about its own product.

For most of this year, that looked like it might finally change. White House officials spent the summer working through a proposal for a national AI standards body, and briefed the largest labs on it in mid-August.

Then three men picked up the phone, and the plan stopped moving.

What a FINRA-style AI regulator would have actually done

The idea came from Demis Hassabis, chief scientist at Alphabet (GOOGL) and chair of its DeepMind unit. He wanted an independent group that could test frontier models before release, staffed with experts capable of probing systems for dangerous cyber, biological and deceptive capabilities.

His model was the Financial Industry Regulatory Authority, or FINRA, the private nonprofit that writes and enforces rules for U.S. brokerages under Securities and Exchange Commission supervision. Member firms fund it, and the SEC reviews its rule changes.

More Artificial Intelligence:

The AI version would have worked the same way. Labs pay in, a common testing standard gets written, and models get measured against it before deployment.

Critics inside the industry made a narrow, honest objection. Voluntary review has a habit of turning into mandatory review, and a model sitting in a testing queue is a model earning nothing.

Why Zuckerberg, Musk, and Huang opposed the AI oversight body

Meta (META) CEO Zuckerberg, SpaceXAI’s Musk, and Nvidia (NVDA) CEO Huang each spoke with Trump separately last month and convinced him to drop the plan, the Wall Street Journal reported Wednesday, Sept. 16. Trump did not move forward, a decision that frustrated some White House officials.

Their stated objection was not that oversight is bad. It was that this particular oversight would hand permanent advantage to OpenAI, Anthropic and Google DeepMind, the three labs already closest to the center of the policy fight, reported Forbes.

Administration officials have since told AI executives that building consensus at the White House is difficult when chief executives can call the president directly, according to the Wall Street Journal.

The public version of the argument landed Tuesday, Sept. 15. Companies have a “strong natural incentive” to keep AI aligned with human interests because nobody uses an agent that ignores them, Zuckerberg wrote in an X post.

Every lab, he added, has “the responsibility and incentive to move at the pace required to train its models safely.”

Huang was blunter at Salesforce’s Dreamforce conference the same day. The choice between safety and speed is “false,” he said, and “we don’t need new laws, we don’t need new regulations,” according to the event video.

What FINRA’s own numbers say about the entrenchment argument

This is where my analysis parts company with most of the coverage I read on this topic. The entrenchment claim got treated as a convenient excuse. I checked it against what the FINRA model actually did to its own industry, and it holds up better than the three CEOs may realize.

FINRA member firms fell roughly 6% to 3,184 in 2025, according to the regulator’s 2026 Industry Snapshot.
Small firms absorbed nearly the entire decline, dropping to 2,832 from 3,048, per the same report.
Large firms held roughly flat near 155, while registered representatives grew to 639,723, according to FINRA.

Fewer firms, more people inside them. That is what a member-funded standards body looks like after 18 years, and FINRA lists regulatory costs among the drivers of that concentration.

So the case Zuckerberg, Musk and Huang made to Trump is defensible. What it does not explain is why the three loudest opponents of a regulatory moat already sit on three of the widest moats in the industry.

The cost side is stranger still. I put FINRA’s roughly $1.5 billion annual budget against Nvidia’s last quarter, and a full year of policing 3,184 brokerages works out to about 1.6% of the $96.2 billion Nvidia booked in three months.

This was never a fight about a membership fee. It was a fight about the calendar, and about who decides when a model is ready.

KENT NISHIMURA / Getty Images

What no AI referee means for your portfolio from here

Blocking a federal body did not leave a clean slate. It left the alternative.

States enacted 109 AI laws and 28 data center laws in the first half of 2026, following 159 AI laws in 2025, according to TechPolicy.Press. Trump’s December executive order created a Justice Department task force whose sole job is challenging those laws.

One standards body with published criteria is something a chief financial officer can budget for. Fifty jurisdictions and a docket of preemption suits is not.

That distinction reaches your account whether you follow AI policy or not. Nvidia closed at $219.34 on Sept. 17, worth roughly $5.3 trillion, and Huang said Thursday he expects to sell twice as many chips next year as this year.

Meta and Nvidia sit among the heaviest weights in the index funds inside most 401(k) plans. So here is the part worth sitting with.

When you own that fund, you are underwriting a safety claim that now has exactly one source behind it, which is the company making it. There is no outside tester and no published standard to check it against when something breaks.

Zuckerberg may be right that market incentives do the job, and Huang may be right that existing law is enough. Neither has to show his work, and that is what actually changed in a phone call last month.

Watch next week. OpenAI’s Sam Altman and Huang are both expected at the White House State Dinner for Chinese President Xi Jinping, and the administration is “open to discussions on avoiding shared risks” with China on AI, Treasury Secretary Scott Bessent told Axios on Wednesday, Sept. 16.

An industry that would not accept a domestic referee is about to talk safety standards with Beijing. Someone in that room is going to notice.

The Arena Media Brands, LLC THESTREET is a registered trademark of TheStreet, Inc.

This story was originally published September 19, 2026 at 1:37 PM.