Senators, Expert Witnesses Compare Notes on AI Safety Concerns - GovTech

Nearly three hours is how long California's Senate Privacy, Digital Technologies and Consumer Protection Committee — a name with more clauses than most statutes — sat and took testimony on the AI frontier. Expert witnesses informed the room that the coming years will see "recursive self-improvement." Not in a paper, not in a leaked memo, but out loud, into a microphone, in a statehouse, without an ambulance idling outside. Three hours of it. Nobody fainted, nobody reached for the emergency brake, and nobody appears to have asked whether a system that rewrites itself should require a licence first.

"Recursive self-improvement" is a marvellous bit of vocabulary — the genteel term for a machine that edits its own code, which every lab has been promising quietly in a blog post while the safety team freshens up its CV. It also converts "the frontier is shifting" from a briefing note into a weather warning. Nothing resembling a bill. No vote, no timeline, no draft language, no subcommittee. What it produced, per GovTech, was notes. Compared. Somewhere in the building, one assumes, Silicon Valley sent slides.

Give them this, though: some senator, unnamed in the report, went on the record with the line that California's AI regulation and governance is "wholly insufficient to meet the moment that is now before us." That is the most honest sentence to leave Sacramento in years, and it was said by a person whose actual job is to remedy the insufficiency in question. Admire the grammar while you can: the moment is before us, the law is behind us, and the distance between them has been quietly reclassified from a gap into a permanent fixture. Three hours of testimony, one admission against interest, and a committee that has now officially compared notes.