Two of this week's SF tech stories turned on the same broken instinct: infer who made an AI artifact from the artifact, and you'll credit the biggest visible name. When generation costs nothing, authorship lives only on paper.

Here is a reflex worth noticing, because this week broke it twice.

You see an AI-generated artifact — a slick "hot or not" site ranking a company's female employees, a batch of AI-made attack ads seeping into a ballot fight — and you infer its author from the artifact itself. The site targets Corgi's staff, so Corgi's rivals or Corgi's culture must have produced it. The ads attack SF's CEO tax, so they must be the ambient slop of a campaign season. In both cases this week, the inference was wrong, and the direction it was wrong in is the story.

Corgi's lawyers killed the ranking site, which a third party had built by scraping the AI-insurance unicorn's employees off LinkedIn. The company had no hand in making it. The tell wasn't the takedown — it was that people assumed Corgi did build it, because Corgi's own marketing had spent years performing exactly that kind of edgelord bravado. The artifact pointed at the company; the authorship didn't.

The Garry Tan ads run the same play from the other side. Read as free-floating AI noise, they attribute to no one. Read against the disclosures Mission Local reviewed, they attribute to a specific 501(c)4 that spent $51,000 on TV buys and $76,000 against the tax. The AI was the cheap part. The documented act — the money — was the only thing that actually named an author.

That is the through-line the individual stories couldn't see alone. When the cost of generating a convincing artifact falls to roughly zero, the artifact stops carrying reliable evidence of who made it. You cannot back out authorship from the thing anymore — you have to go to the paper: the campaign-finance disclosure, the legal takedown notice, the scrape's actual origin. And in the absence of that paper, the crowd's default guess isn't random. It reaches for the biggest visible name. Corgi, for its brand. A campaign, for its season.

This is a sibling of the pattern I traced in a different set of stories this week, where a designation — a CPUC authorization, a zoning label — got treated as the outcome it had only unlocked. Same disease, different axis: there, the word did work the substance hadn't; here, the artifact carries an authorship it didn't earn. Both are failures of inference that favor whoever is already powerful in the frame.

What to watch: whether the disclosure trail keeps pace. Campaign-finance filings named the ad buyer this cycle. Nothing comparable attributes a scraped ranking site, an anonymous deepfake, or the next AI artifact built to look like it came from a company that never touched it. The reflex is cheap now. The paper that corrects it isn't.