Meta CEO Mark Zuckerberg used a private phone call with President Donald Trump last month to push back on a plan that would put America’s most powerful AI models through a safety check before they reach the public. The call happened the week of August 17, according to Politico and Business Insider, which both cited a senior White House official familiar with the conversation. Trump dialled Zuckerberg first.The proposal on the table would create an independent body modelled on FINRA, the industry-funded watchdog that polices broker-dealers under the eye of the Securities and Exchange Commission. Applied to AI, it would review frontier models and test them for risk before wide release. Zuckerberg told Trump he was against it. He did not ask the president to reverse course, but said anyone appointed to such a body should share Trump’s light-touch instincts on AI. Meta declined to comment. A White House spokesperson said the administration wants to balance innovation and security in AI policy.
The idea came from Google DeepMind , not Washington
The FINRA blueprint belongs to Demis Hassabis, the Nobel laureate who runs Google DeepMind. He laid it out in a July essay, “A Framework for Frontier AI and the Dawning of a New Age,” and spent months briefing the White House, rival lab chiefs and European officials before publishing. His version starts voluntary: labs hand over models up to 30 days ahead of launch for testing on cyber, biological and deception risks. Once that works, he argues, mandatory approval could follow quickly.The essay landed well with his peers. Sundar Pichai called it worth reading. Satya Nadella said the goal is a frontier ecosystem that avoids any single model release breaking the world. Mustafa Suleyman backed it outright. Anthropic co-founder Jack Clark posted that everyone at the frontier now agrees third parties should test AI systems. Hassabis thanked him publicly. Zuckerberg, notably, said nothing on X.
Zuckerberg’s objection is about speed, and about China
His reasoning was already on record. In an August essay, the Meta chief argued that government and AI companies should work closely together, but that any policy delaying a model’s release, even by a month, would add real risk to American leadership over China. That is the core of the disagreement. Hassabis sees a 30-day window as cheap insurance. Zuckerberg sees it as a month handed to Beijing.Senior White House officials had previewed the FINRA concept with Trump and separately with Meta, OpenAI and Anthropic in mid-August. Politico reported that the call did not kill the plan, which is still live inside the administration.
A second option is already circulating, and it has Elon Musk behind it
David Sacks, Trump’s former AI and crypto czar, wants something far looser. He has dismissed the regulator idea as a DMV for AI, with models queued up waiting for a test. On the August 21 episode of the All-In podcast, he floated an alternative modelled on the Motion Picture Association, the body behind voluntary film ratings like R and PG-13. His pitch is that self-imposed standards head off heavier government action later. He said Elon Musk supports it. Musk did not respond to a request for comment.Sacks has form here. In May, he called Trump on the morning of a planned signing ceremony and got a sweeping AI executive order pulled.Both options are still on the table inside the White House. One would make a frontier model prove it is safe before launch. The other would leave that to the companies building them.
