Three Tech Chiefs Reportedly Talked Trump Out of an Industry-Funded AI Watchdog
Zuckerberg, Musk and Huang each reached the president directly and helped stop a FINRA-style oversight body pitched by Google DeepMind’s Demis Hassabis, according to a Wall Street Journal report circulating Thursday.
Three of the most powerful figures in American technology reached President Donald Trump directly in recent weeks and helped stop a plan for an industry-funded artificial intelligence regulator, according to a Wall Street Journal report that circulated widely on Thursday.
The executives were Meta’s Mark Zuckerberg, Elon Musk of SpaceX and xAI, and Nvidia’s Jensen Huang. Each spoke with Trump separately, the Journal reported, and after those conversations the White House did not advance the proposed oversight body.
The striking part is where the idea came from. It was not a bill from Congress or a rule from a federal agency. It was a proposal from inside the industry itself.
What was on the table
The plan had been pitched to White House officials by Demis Hassabis, the chief scientist of Google DeepMind, who chairs the company’s AI unit. Modeled on FINRA, the private body that oversees US brokerage firms under government supervision, it would have created an industry-funded standards and testing organisation to evaluate frontier AI models for cybersecurity, biological and deception risks before public release. Reporting indicates the checks would have begun as voluntary, with the possibility of becoming mandatory later, under government oversight.
Supporters argued it would set a common safety floor without waiting for slow legislation, and keep American labs in the lead while lowering the odds that one company ships a dangerous system and the whole industry pays for it. Some White House officials and several AI executives were open to it. Zuckerberg, Musk and Huang were not.
Why the three objected
People familiar with the talks told the Journal the core objection was about power, not paperwork. The three reportedly feared the body would entrench the influence of the labs already at the centre of the frontier debate, OpenAI, Anthropic and Google DeepMind, turning a safety mechanism into what critics call a regulatory moat. Who would be appointed, who would design the tests, and who could slow a rival’s release became the real fight.
That is a telling detail. The proposed watchdog was not a government agency with subpoena power. It was an industry-funded body. The three chief executives still treated it as a threat to their freedom to move. According to the Journal, White House aides have since told other AI executives that reaching consensus inside the administration is difficult, because those opposed to new rules simply call the president.
The public case
The private lobbying coincided with a public argument against slowing down. In a lengthy post on X, Zuckerberg argued that companies already have a “strong natural incentive” to keep AI aligned with users, since people will not keep using agents that ignore their instructions. Each lab, he wrote, has both the responsibility and the incentive to train its models safely on its own. He left one opening: Meta is willing to work with independent evaluators, a stance he said he shares with OpenAI and Anthropic.
Huang was blunter. Speaking at Salesforce’s Dreamforce conference, he rejected the idea that safety and speed are a trade-off, saying a company can pursue both and that new laws and regulations are not needed. If a firm is not confident a product is safe, he argued, it should not ship it, and the market will punish those that do.
Musk’s public posture this month has been mixed. He has mocked the most alarmist safety warnings while also saying, separately, that Anthropic chief executive Dario Amodei was right to call for better coordination and pacing. Unlike OpenAI and Anthropic, Musk’s AI company has not published a formal plan to slow or jointly test frontier releases.
Trump’s position has been consistent. He has dismissed existential warnings about AI and argued that the main safeguard the country needs is a strong and capable president, a line he has repeated publicly. During a live appearance with Huang, he warned that opponents of AI and data-centre construction were playing into the hands of skeptics and of China.
A White House that is not of one mind
The Journal’s account describes an ongoing tug-of-war rather than a settled policy. On the lighter-touch side are figures including White House AI and crypto adviser David Sacks, who has floated a softer alternative closer to a voluntary ratings system, one that would label AI systems rather than gate their release. On the other side, officials responsible for the economy and for cyber security have pushed for more scrutiny of shared risks, and the administration has signalled it remains open to discussion even as the president dismisses the loudest alarms.
It is not the first time this circle has stalled a tighter rule. Earlier reporting this year said a draft executive order that would have required companies to submit models to national security agencies for pre-release testing was shelved after similar objections.
What the fight is really about
Strip away the branding and three questions remain:
- Who sets the rules for the most capable models, the labs that build them or a body that can slow a launch?
- Does an industry-funded watchdog protect the public, or does it become a club that protects the incumbents?
- If the United States refuses a shared testing regime, does that keep it ahead of China, or simply leave accidents and misuse to be cleaned up after the fact?
For now the answer is clear enough. There will be no FINRA for frontier AI. The companies will keep building, the White House will keep taking their calls, and the argument over whether that is confidence or capture is only getting louder.