Powerful Dem trio demands faster AI incident disclosures - Politico

Oh my goodness, *forty-eight hours*! ✨ Frank Pallone, Zoe Lofgren and Jamie Raskin have written to five frontier AI companies — Anthropic, Google, Meta, OpenAI and the delightfully mysterious SpaceXSI — asking to be told about major AI safety incidents within two whole days, which is *two thousand, eight hundred and eighty minutes* of thoughtful, unhurried transparency. Most people don't reply to a dinner invitation that fast! Congress has essentially been handed a calendar with room to breathe on it, and what could possibly be more considerate than that?

And the second request — an account of "all known safety incidents to date" — is honestly the loveliest thing anyone has asked a lab in years. It's a show-and-tell! It's a scrapbook! Five companies are being invited to sit down and reflect on everything they've learned, which is precisely the kind of self-awareness the industry keeps promising us it has. Every one of them gets to make a list, and lists are *lovely*. Yes, the letters don't define the word "incident", but that just leaves space — space for interpretation, space for nuance, space for a really generous reading in which the model was simply being exploratory. ✨

What I adore most is the format: letters. Not subpoenas — *letters*! Which means nobody is being accused of anything, nobody's lawyers are having a bad week, and every one of the five recipients gets the warm glow of being asked politely rather than told. Our own recent fortnight of AI coverage has been one long incident after another, and here at last is somebody saying: could you perhaps jot it down? In a document? For Congress? Within 48 hours? The paperwork for the paperwork is coming, it will have a lovely font, and it is going to be *magnificent*. ✨