What's the point of 6,500 words?
Mark Zuckerberg published about 6,500 words yesterday arguing that the real danger in AI is a handful of companies controlling it. He made that same argument thirteen days earlier, in an op-ed in The Wall Street Journal with the headline "The AI Future Is for Everyone." But they arent' the same document; the op-ed framed the argument and the manifesto is attempting to carry the promises.
And he's hardly alone. The CEO manifesto is suddenly its own genre, with new entries dropping almost weekly.
Business Insider asked the question I had this morning, which is: is anyone reading these things? No. Maybe competitors, maybe a few venture capitalists. And that's fine, because the op-ed and the manifesto have different jobs. The op-ed was built to be read, it's short, you can find it in a paper critics already subscribe to, and it's on the opinion page a Senate staffer skims over breakfast. The 6,500-word version was built to be quoted. Think of it like the terms and conditions on your phone: nobody reads them, but they exist so someone can point to a paragraph later. A trade group footnotes it, a lobbyist slides it across a table, a policy aide lifts the one line she needs.
And there's a reason it's happening right now.
The Trump administration is about to write the rules for the most powerful AI systems around what gets checked, by whom, before it's released. So now everyone with money on the table is suddenly publishing their opinion; Axios counted five in fifteen days. Demis Hassabis on July 14, proposing a safety-testing group the industry would pay for but the government would oversee. Jensen Huang on July 24, with a letter arguing AI models should stay free for anyone to download and build on; Meta and Microsoft signed on day one. Dario Amodei on July 27, wanting mandatory safety testing and tighter limits on who gets advanced chips. Zuckerberg's op-ed on July 28. And a petition that same day from more than 1,200 people who work inside these labs, asking Washington for a way to slow things down on purpose. Every one of these things is trying to shape which risks the new rules take seriously so the real audience for Zuckerberg's essay is roughly the size of a congressional subcommittee.
The Meta manifesto also has a fresh batch of promises, and they come in two piles. The real ones: free AI tools for billions of people, a private mode where even Meta can't see your information, a return to releasing its models publicly (which the company says is coming, not something it's done) and a fund for the towns where Meta builds data centers. Then there's the other pile: somebody 'should' share early versions of models with the government for example. It says AI labs "should" do this, not Meta committing to anything.
More of these manifestos are likely coming. The rule isn't written yet, and until it is, every CEO with money on the table is watching their peers publish... because nobody wants to be the one who stayed quiet while the risks got defined by somebody else. So when the next manifesto drops, consider skipping the debate about whether anyone reads it and ask the two questions that tell you what it is:
who are the ten people this was written for, and
what is it asking them to do?
If you can't name the audience, it's marketing. If you can, it's lobbying and now you know which rule to go watch. And if your own CEO catches the bug, you know the drill from above: the one-page version for the people who decide, and the list of promises sorted into "we will" and "somebody should," with a name next to each "we will."