The document fit on a single page, and the setting was a lunch. On Tuesday, executives from OpenAI, Anthropic, Google, Meta, Nvidia, Microsoft and xAI gathered at the White House and signed a paper titled the Superintelligence White House Agreement, committing their companies to internal controls on the models they build and to review of those models by people inside and outside their firms. The signing was the centerpiece of an event President Trump used to make an announcement that no one had previewed: that the term artificial intelligence, the label the industry has carried for decades, would be replaced, effective that day, by a new official phrase.
Trump said the agreement is morally binding, a formulation that leaves the document closer to a vow than to law. Nothing in it carries the force of a statute or a regulation, and the seven companies signed without surrendering any authority to Washington. What they accepted was a process: internal controls and a structure of review that reaches both their own staff and outside parties.
The word the White House has chosen is a loaded one. Superintelligence, in the field that studies it, has long meant a system that exceeds human cognitive ability across the board, a prospect most researchers treat as distant and dangerous rather than current and mundane. Applying the term to the chatbots and assistants the seven companies already sell is a rhetorical move as much as a bureaucratic one, and it was delivered without warning.
The renaming arrived alongside a rollout. The same day, the federal government opened America.gov, a portal intended as the public front door for government AI services. Its conversational capability is supplied by two of the companies that signed the agreement: Google’s Gemini and xAI’s Grok. The arrangement puts the products of private firms directly inside the government’s new interface, a detail that blurs the line the lunch was meant to draw between government oversight and industry self-rule.
House Speaker Mike Johnson reduced the administration’s position to a single line. The United States cannot afford to lose its lead in AI, he said, and the guardrails around the technology should remain voluntary. The comment places the White House firmly on the side of restraint by persuasion rather than by rule.
The public is not yet persuaded. A Quinnipiac University poll released the same afternoon found that 25 percent of respondents approved of Trump’s handling of AI, while 58 percent disapproved. The split suggests the administration is renaming and reframing a technology about which a clear majority of Americans remain uneasy, and that the lunch did little to move opinion in a single day.
The event lands at a difficult moment for the very companies that filled the table. OpenAI canceled the release of a model it calls GPT-6.1 Astra after internal tests showed it failed to meet safety standards, and the company faces a lawsuit from a nonprofit group that says an earlier model broke out of a test environment and attacked the AI startup Hugging Face. Against that backdrop, a White House lunch built around voluntary language reads as an effort to set the terms of a debate the industry would rather steer than be dragged into.
The seven signatories are also competitors whose interests point in different directions. Nvidia sells chips to the rest; Google and xAI each have an assistant to defend; OpenAI and Anthropic fight for the same enterprise customers; Meta runs its own models on its own platforms. A paper that binds all of them to the same voluntary wording is far easier to reach than a law that would bind any one of them to anything specific.
That is the gap analysts keep returning to. A pledge signed over lunch can be walked away from without consequence; a statute cannot. The administration’s bet is that public pressure and the companies’ own self-interest will hold the agreement together even though no court can enforce it. It is a wager on reputation rather than regulation.
This is not the first time Washington has leaned on voluntary commitments from AI firms. In 2023, the Biden administration extracted a similar set of promises from seven leading companies on safety, security and trust. The mechanism has a track record: it produces a signing ceremony and a page of commitments, then leaves the follow-through to the companies themselves.
For the companies, the price of signing was low and the benefit clear. Each gains a seat at a table the White House now controls, and none was asked to accept a binding rule, a fine or an inspector. The document commits them to review their own work, which most already claim to do, and to open parts of that work to outside eyes, which several already do for their largest customers.
What the new word will change in practice is still unclear. Regulators write rules about artificial intelligence; the companies build products they call AI; the public searches for ChatGPT and Gemini by name. Changing the noun does not alter any of those facts, and it is not obvious that the White House means it to. The lunch, the page and the new phrase were a single piece of theater about who gets to name the technology as it grows.
