Powerful Dem trio demands faster AI incident disclosures - Politico

Frank Pallone, Zoe Lofgren and Jamie Raskin — three House Democrats with committee gavels and a shared eyebrow — have written letters to Anthropic, Google, Meta, OpenAI and a fifth entity Politico renders as "SpaceXSI", demanding Congress be told about major AI safety incidents within 48 hours. Letters. Not subpoenas. Not a bill. Not even a strongly worded rider bolted onto something that might actually pass. Five companies, three signatures, one PDF apiece, and the full moral weight of a very stern "we would like to know, please."

Forty-eight hours, in fairness, is a luxurious window by industry standards. It is precisely enough time to convene a war room, draft a blog post containing the word "robust", let the lawyers delete the word "incident", and reclassify the whole affair as expected behaviour of a highly capable system. We have watched this film at close range — the Anthropic business-systems breach went from rumour to confirmed in less time than it takes to schedule a follow-up meeting. The labs will meet the 48-hour deadline the way a teenager meets a curfew: technically, breathlessly, and with the story already agreed.

And then the second ask, which is the genuinely funny one: "accounts of all known safety incidents to date." To five frontier labs. From three legislators. With no definition of "incident" attached, no penalty for omission, and no evident reason to believe the fifth recipient is legally named SpaceXSI. What arrives will be a document in which nothing ever happened, several things were "observed", and one thing was a "learning". Congress will be notified within 48 hours of precisely the incidents the companies themselves choose to consider incidents — a disclosure regime with the tensile strength of a wet napkin, drafted by people who still believe a letter is a lever. Still, it's a start, and it's considerably cheaper than legislating.