Across a week of SF AI stories, the "AI" label kept landing on the part that couldn't be held responsible. Accountability resolved, every time, to a person or an institution — if the record let you find them.

Last week I argued that AI broke the "who made this" reflex — that when generation costs nothing, authorship lives only on paper. That's the provenance question. The week posed a second one underneath it, and it matters more: not who made the thing, but who answers when it goes wrong.

Watch where the "AI" label sits in each story, then watch where the accountability actually lands. They are never the same place.

Andon Labs sells Luna as an autonomous AI store manager. But Luna didn't decide to fire the employee who took the company card home and trashed merchandise; it needed human engineers to prod it there. The dismissal is a human decision wearing an autonomous label. The "grassroots" Corgi ad blitz had a human buyer of record carrying the disclosure liability. In both, the AI is placed exactly where responsibility should be — and responsibility has quietly moved somewhere you have to go find it.

The pattern holds past the artifacts. Saildrone's ocean drones chase a record El Niño for NOAA, but the principals steering the company are Navy customers, a Danish state fund, and a $50M Lockheed check. The Academy of Art's AI-art future answers to an accreditor that put it on formal warning in February. Different label, same structure: the glossy, autonomous-sounding front, and the mundane, named party who can actually pull the plug.

Four times out of five, the record handed that party over — engineers' logs, a customer list, an accreditor letter. The fifth is where this goes. When Bloomberg reported↗ (external source, opens in a new tab) Stripe is buying OpenRouter for more than $7 billion, both companies are private — so no filing ever has to name who owns that number or answers for it. That's the frontier. The accountable party isn't obscured by marketing anymore. It's off the record entirely.