A White House meeting on AI regulation ended with tech leaders agreeing on an AI "constitution." Stuart Russell, president of the International Association for Safe and Ethical AI, joins "The Daily Report" to break it down.
Page A10From§Eachthe milkman edition — 1 October 2026
Trump revives Biden's AI self-policing pledge days after a Senate warning on nuclear command and control.
As it ran on the front
Start on the 30th of August, when Google Maps renamed Lake Ontario "Lake America." It was a proof of concept. Not long after, the government's own AI chatbot, the one built to answer citizens' questions, started contradicting President Trump's claims — and then its answers were edited until it stopped. Correction, applied until the record matched the man, not the facts.
On the 21st of September, a defense writer described an AI glitch that had nearly started a war, and warned that a catastrophic technology was being treated as an afterthought — something nuclear command and control was never allowed to be. The next day, a UN-backed scientific panel told world leaders the safeguards that already exist are "unraveling."…
…(cont) Three days later, Bill Gates — co-founder of Microsoft — told reporters he does not share the industry's confidence in its own self-regulation, and prefers a hands-on approach from government instead. That is the dissent that arrived before the room even convened.
The room convened anyway. Trump's answer was a voluntary pledge: tech executives signing what CBS described as an AI "constitution," built on the same design the same companies ran under President Biden — write the rules, grade the homework, same as before. The New York Times asked the only honest question in its headline — is it enough — the same day its reporting found mounting examples that the companies are already failing to put safeguards on the technology they are now pledging to police.
That Wednesday, a Senate hearing took up what happens when an AI model goes rogue. The same day, on CBS's "The Takeout," John-Clark Levin, head of research at Kurzweil Technologies, walked through what the hearing had explored and named the scenario directly: a cyberattack on the systems that run nuclear command and control. Not a novelist's premise. A possibility serious enough for a Senate committee to spend an afternoon on, in the same week the industry it concerns put its signature on a pledge to watch itself.
By Thursday, the government chatbot had stopped answering some political questions altogether — not edited now, silent. The same day, Stuart Russell, president of the International Association for Safe and Ethical AI, told CBS he does not believe the companies' pledge will hold. The same day, Jordan Klepper told "The Daily Show": "Sure, we don't need laws! I'm sure we could trust these guys." A UN panel, a Microsoft co-founder, an AI safety president, and a comedy writer reached the same verdict this month. Only one of them is paid to be funny about it.
“They ran this exact self-policing play under Biden, and what it produced was a government chatbot that had to be edited until it agreed with the boss. Same week a Senate hearing takes up a model going rogue, a researcher goes on CBS and names the actual nightmare — a cyberattack on nuclear command and control — and the fix on the table has zero inspectors written into it. That's not a gap in the plan. That is the plan.”
“Look, they signed a constitution — that's more structure than Biden ever got out of these companies, frankly, voluntary or not. The nuclear command thing, that's a researcher on a cable hit talking about a hearing, not some imminent — okay, a Senate committee did spend an afternoon on it, fine, but nothing's actually happened happened. And the chatbot not answering political questions anymore, that's just restraint. Did I say restraint? That's not the word I meant.”
Earlier in this story
- Trump's AI chatbot disputes his claims, then stops disputing them. By Sal · the wee small hours edition, 1 October 2026
- AI industry polices itself again, same plan Biden tried, same week Senate hears it could trigger nuclear war. By Sal · the midnight edition, 1 October 2026
- Trump's AI pledge is already 'entirely unenforceable,' and the industry's own money just walked out on it. By Sal · the late evening edition, 30 September 2026
- Congress passed no AI law, so no senator could say who stops a nuclear AI glitch. By Sal · the early evening edition, 30 September 2026
- AI CEOs sign 'morally binding' pledge; FTC opens investigation into signers days later. By Sal · the supper edition, 30 September 2026
- Senate Democrats kill data center bill as too weak, their candidate cashes its industry's checks. By Lou · the matinee edition, 30 September 2026
The receipts
The US government's AI chatbot has stopped answering some political questions after it drew attention for contradicting some of President Donald Trump's contentious political claims.
A Senate hearing on Wednesday explored the increasing fears about AI models going rogue and causing chaos. John-Clark Levin, head of research for Kurzweil Technologies and former head of research for Google's chief futurist, joins "The Takeout" to discuss.
President Trump’s plan to let A.I. companies police themselves comes amid mounting examples that they are failing to put safeguards on the technology.
“Sure, we don’t need laws! I’m sure we could trust these guys,” Jordan Klepper said of technology executives on “The Daily Show,”
· Bill Gates disagrees with AI leaders on self-regulation from the morgue, 25 Sep 2026
· UN panel calls for guardrails on AI as current safeguards are 'unraveling' from the morgue, 22 Sep 2026
· An AI glitch almost just started a war. Next time, we won’t be so lucky from the morgue, 21 Sep 2026
This page is a back-issue: the story as it ran, receipts as they were. The current edition is at the front. The byline is a pen name for a column drafted by a machine and checked by the editor: how this is made.