Imagine a restaurant that fails a health inspection. Instead of scrubbing the kitchen, management changes the sign out front to something grander, prints a new menu, and asks the chefs to sign a pledge promising they care deeply about hygiene. No inspector, no fines, no follow-up visit. Just a framed piece of paper near the host stand. You would still think twice about the shrimp.
That is roughly where we are with the White House Accord on Super Intelligence, a document signed by President Donald Trump and the leaders of six of the top U.S. AI companies. Alongside it came an executive order replacing “artificial intelligence” with “super intelligence,” or SI, across U.S. government communications. Two moves, one week, aimed squarely at an industry with a trust problem. Neither move does much about the trust problem.
A pledge with no teeth and a typo
Trump called the accord “morally binding.” That phrase deserves attention, because it is the tell. Contracts are legally binding. Regulations are legally binding. “Morally binding” is what you say when you want the weight of an obligation without any of the mechanics. The accord is voluntary. It lacks enforceability. It lacks clear lines of responsibility, which matters enormously in a field where the interesting question is never “do you intend to be safe” but “who is accountable when the system fails.”
The Council on Foreign Relations called the pact toothless, which is about as blunt as foreign policy institutions get. On TechCrunch’s Equity podcast, Kirsten Korosec, Sean O’Kane, and their co-host were collectively skeptical about whether any of this changes the industry’s direction. Korosec suggested the real significance was the meeting itself, which is a generous reading, and probably the correct one.
Then there is the detail that will outlive every talking point in the document: the agreement misspelled the United States. Not a signatory’s name, not a technical term. The country. That is not a nitpick about proofreading. It tells you how fast this thing was assembled and how little scrutiny it survived before signatures went on it. Documents that get lawyered for weeks do not have that error. Documents produced for a photograph do.
Why the rebrand is the weirder half
The accord at least gestures at a real concern. The renaming is harder to explain. “Artificial intelligence” is an imprecise term, sure, but it is the term, with decades of research, legislation, and public understanding attached to it. “Super intelligence” is not a correction. It is an upgrade in marketing language applied to the same systems doing the same things with the same failure modes.
If anything, it works against the stated goal. The public image problem with AI is not that the technology sounds too modest. People are worried about job displacement, slop content, surveillance, and tools that confidently make things up. Telling them the thing they distrust is actually super intelligent does not calm anyone. It raises the stakes of every complaint they already had.
For those of us who test these tools for a living, the gap between naming and behavior is familiar territory. Every agent platform I review has a product page promising autonomy and reasoning. What I actually get, more often than not, is something that loops on a simple task or silently fails a step and reports success. The vocabulary has always run ahead of the capability. Now the vocabulary is running ahead at the federal level.
What would actually move the needle
None of this means the meeting was pointless. Getting the president and six major AI company leaders in a room and on the same document is not nothing. It establishes that safety is a topic the industry has to show up for, in public, with names attached. That is a floor, not a ceiling.
The useful version of this would involve a few things the current accord does not have:
- Specific obligations tied to specific actors, so responsibility is traceable when something breaks
- Some mechanism of verification that does not depend on companies grading their own work
- Consequences that exist outside of reputational discomfort
- Enough drafting time that the document spells the country correctly
The CFR framing is the right one: toothless now, with a path forward. A voluntary framework can become the skeleton of something enforceable, but only if someone treats it as a first draft rather than a finished product.
My read, as someone who spends most of his time finding out which AI products do less than they claim, is that this follows the same pattern. A confident name, an impressive signing, and nothing underneath that would change how a single model gets built or deployed next quarter. The image problem is downstream of a performance and accountability problem. You do not fix that with a new word and a pledge nobody has to keep.
🕒 Published: