Comment les patrons de Meta et Nvidia ont œuvré en coulisses pour l’accord de la Maison-Blanche sur l’IA

The CEOs of Meta, Mark Zuckerberg, and Nvidia, Jensen Huang, have been behind the scenes working on voluntary agreements on AI with the White House, according to five unnamed sources familiar with the closed-door meeting at the White House on Tuesday. During the lunch, Zuckerberg and Huang presented these AI agreements to top White House officials and several major names in tech, the sources said.

Huang and Trump administration officials contributed to drafting the text before the meeting, two of the sources indicated. These five sources spoke on condition of anonymity as they were not authorized to reveal the details of the discussions. In this voluntary agreement, signatories would recognize the importance of putting safeguards in place to mitigate potential risks associated with advanced AI models.

However, due to the lack of a clear enforcement mechanism, it is uncertain what consequences they might face if they fail to adhere to their commitments. This episode highlights the growing influence of Zuckerberg and Huang on US policy regarding AI, as the administration continues to prioritize a non-interventionist approach to regulating this technology.

Congress has yet to pass federal legislation to govern AI. Meta declined to comment, and Nvidia and the White House did not respond immediately to requests for comment. The idea originated during a state dinner hosted in honor of Chinese President Xi Jinping, where Zuckerberg reportedly introduced it to Mike Johnson, according to one of the sources mentioned earlier.

Soon after, Zuckerberg exchanged with Huang, who then helped rally the leaders of the major AI labs to the project, two of the sources said. Between that state dinner and Tuesday's lunch, leaders communicated regularly to finalize the document. Earlier this week, Semafor reported on some details of the document's drafting. The agreements commit companies to "implement rigorous internal controls" to monitor AI model capabilities, collaborate with external auditors, and create independent committees to review their reports.

Some Democrats and AI safety advocates have criticized the agreements as unenforceable and relying mainly on mutual agreement. Zuckerberg told journalists at the White House on Tuesday that "this is just the beginning, an agreement to which the entire industry could adhere." AI regulation is a contentious issue Donald Trump has been reluctant to impose new safeguards on this technology, considering existing laws sufficient to mitigate potential risks.

"The only control or 'safeguards' AI needs is a strong and intelligent PRESIDENT (high IQ!), and the United States has one," he posted on his Truth Social social media platform in mid-September. While the White House deliberated on the best way to regulate AI, Zuckerberg advocated for a less interventionist approach. In a phone conversation with Trump in August, Zuckerberg opposed a proposal to create an independent regulatory body inspired by the Financial Industry Regulatory Authority for financial markets.

Nvidia unveiled a new AI security system on Monday, designed to prevent hacking by autonomous AI agents, similar to recent high-profile incidents. Other top tech executives, including Dario Amodei of Anthropic and Sam Altman of OpenAI, have called for more government intervention to regulate AI and slow its development. Zuckerberg and Huang largely dismissed concerns that AI could pose an existential risk to humanity.

This story was originally published by POLITICO in English and then edited in French by Jean-Christophe Catalon.