Multiple people familiar with the discussions told Semafor that the concept behind the pact President Donald Trump unveiled with AI executives on Tuesday began with a conversation between Meta CEO Mark Zuckerberg and House Speaker Mike Johnson at last week’s state dinner.
According to a source, Zuckerberg — who sat next to Mike Johnson at the dinner for Chinese leader Xi Jinping — raised regulatory concerns about artificial intelligence. He then spoke with Nvidia CEO Jensen Huang about drafting a set of principles that tech executives could present at their upcoming meeting with Trump.
The source said Zuckerberg circulated a draft prior to Tuesday’s meeting between Trump and the AI industry leaders. Jensen Huang helped gather support for the draft among the administration’s guests, which included executives from OpenAI, Anthropic, xAI and others. President Trump ultimately released the document as the “White House Accord on Super Intelligence,” a roughly 300-word statement laying out broad principles for industry self-regulation.
The administration and the executives who attended described the accord as a historic consensus among competing companies and a sign that the industry intends to act responsibly. It remains unclear, however, whether this gesture will alleviate widespread public concerns about AI safety.
Two additional sources confirmed Zuckerberg’s central role in shaping the document, suggesting his cultivation of relationships in Washington has produced influence comparable to high-profile figures such as Elon Musk and Jensen Huang. Meta did not comment for this story, and the White House did not respond to a request for comment.
What the accord says
The agreement, signed by Meta, OpenAI, Anthropic, xAI, Google and NVIDIA, commits signatories to implement “robust internal controls” and external audits intended to ensure the safety of frontier AI models. Some models would undergo voluntary government clearance under a framework established by a Trump-era AI order, though that framework has not been publicly released.
At just over 300 words, the accord does not provide detailed descriptions of what industry-run safety controls would look like in practice. It also signals that federal AI regulation may be necessary in the future.
Zuckerberg called the accord “good steps forward” on Tuesday. Speaking to reporters outside the White House, he said: “The idea isn’t that this is the only thing that we will ever do. It is that this is a start and an accord that the whole industry could come to.” He appeared alongside President Trump, House Speaker Mike Johnson and other attendees.
Why this matters
The episode illustrates how informal conversations and personal relationships can translate into a formal industry-government statement. While the signatures of major AI companies could indicate willingness to cooperate on safety, the accord’s brevity and lack of operational detail leave open questions about its effectiveness in addressing public safety concerns.
As background, The Wall Street Journal has reported that Meta is seeking to expand its presence in the business AI market with a new Muse agent.
