Mark Zuckerberg told President Trump that a FINRA-style AI regulator was the wrong answer for America's AI race. The proposal is still alive, but Meta's chief executive has already put his thumb on the scale.
You don't need a lobbyist when you've got the president's ear. Business Insider reported on September 3 that Trump spoke with Zuckerberg during the week of August 17 about a White House proposal to create a national AI regulator, and Zuckerberg told him he opposed it. One person cited by the outlet said Trump placed the call.
The proposal has a real author behind it: Demis Hassabis, the Nobel Prize-winning co-founder and chief executive of Google DeepMind. In July, Axios reported that Hassabis called for a U.S.-led standards body modeled on FINRA, the Financial Industry Regulatory Authority that polices Wall Street brokers under SEC supervision. The idea is not an FDA for every chatbot. It would focus on frontier models, with labs initially sharing systems before release for testing on cyber, biological and deception risks.
That distinction matters. FINRA is funded by the financial industry, but it works under the SEC. An AI version would give Washington a standing mechanism for pre-release review without building a brand-new federal agency from scratch. Hassabis wants it running before year-end, Axios reported. Some White House officials like the model. Zuckerberg does not.
His objection isn't new. Zuckerberg spent more than 6,000 words making a broader version of the same case in August, in an essay titled "The Future Is for Everyone," published the same day Meta released Muse Glimmer, a 30-billion-parameter open-weight model built to run local AI agents on consumer hardware. His argument is blunt: any policy that slows American model releases, even by a month, gives foreign rivals more room to catch up.
Zuckerberg bets billions of personal AI agents on Meta's most expensive quarter ever Meta CEO Mark Zuckerberg predicted billions of personal AI agents within five years on the same Q2 2026 earnings call where profit fell 14%, free cash flow dropped 91%, and the stock tumbled nearly 10% after hours. The gap between the vision and the current financials is the real story. - personal AI agents for consumers - Meta expensive quarter cash flow concerns
Here's the thing. Zuckerberg is not only arguing about safety rules. He is defending Meta's open-weight strategy at the exact moment rivals such as OpenAI and Anthropic are more comfortable with tighter controls on the most advanced systems. If you run closed models, a formal review regime can become a moat. If you want developers down and adapting your models, it can become a gate.
Business Insider reported that Zuckerberg did not ask Trump to kill the FINRA-style body outright. Instead, the outlet said he argued that anyone appointed to such a body should reflect Trump's own light-touch approach to AI. That's a smaller ask, but it may be the sharper one. Regulators can last. Appointees set the tone.
The idea survived the call anyway. Business Insider reported that the White House is still considering the FINRA-style regulator alongside another option, a voluntary industry group modeled on the Motion Picture Association rating system. David Sacks, Trump's former AI and crypto czar, has pushed that looser model and called a government regulator a "DMV for AI" on the August 21 episode of the All-In podcast.
The split is no longer theoretical #
Not everyone in Silicon Valley is standing with Zuckerberg. Axios reported in July that Hassabis, OpenAI's Sam Altman and Anthropic's Dario Amodei had each published proposals pointing toward outside scrutiny of frontier models, even if they disagreed on the exact machinery. Amodei has argued for an FAA-style agency with the power to block releases. Hassabis prefers the FINRA structure. Altman has pushed for an international certification system.
Zuckerberg's path is different. Keep the releases moving, keep the models open where possible, and use closer cooperation with government instead of a fixed approval line. That sounds cleaner if you're Meta. It sounds much riskier if you believe the next generation of models could help automate cyberattacks or lower the barrier to biological misuse.
There is a business fight underneath the safety language. OpenAI, Google DeepMind and Anthropic already have the lawyers, security teams and federal relationships to survive a certification regime. Smaller labs and open-source developers would have a harder time. A regulator built to control the frontier could still end up protecting the incumbents that helped shape it.
Zuckerberg has done this before #
This is the second time this year. Pressure from Silicon Valley has changed the shape of Trump's AI policy. The Washington Post reported in May that last-minute calls from industry leaders, including Musk and Zuckerberg, helped block an expected executive order that would have created a government review process for advanced models before public release. Semafor later reported that Trump signed a lighter version on June 2, with the review period reduced to 30 days.
Meta and Microsoft walk into earnings week with $145 billion question marks hanging over them
Meta reports July 29 with a $145 billion AI spending plan and a stock already under pressure. Microsoft reports the same evening with a $37 billion AI revenue run rate but capex expected to exceed $40 billion in Q4 alone. Together with Alphabet and Amazon, the four hyperscalers are on pace to spend $725 billion on AI infrastructure in 2026, and... - meta capital spending justification earnings - microsoft ai investment return proof
Same script, twice this year. Tech executives argue that friction with China is the real danger, and federal AI policy bends toward speed.
That doesn't make Zuckerberg wrong. A slow, clumsy approval regime could hurt American labs while Chinese open-source models keep improving. But it also doesn't make him neutral. Meta has a direct commercial interest in avoiding a regulator that can slow or second-guess open-weight releases.
The proposal is still on the table. So is Zuckerberg's number.
Also read: Ray Kurzweil joins Subsense, the brain interface startup you snort • OpenAI Says GPT-6 Astra Marks the Start of the AGI Era • Conphere Puts AI Agents Behind the Curtain of Event Planning
Founder discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.