Bakery prospectus nearly as scary as Anthropic's S-1, trial finds
Anthropic tested its pre-IPO filing on readers against a dummy document about a bakery, and the bakery was nearly as scary.

"...and the placebo group was just as afraid of AI, which is the result I will be thinking about for the rest of my life," said Dr. Marguerite Fenn, who ran the trial.
Anthropic is getting ready to go public. It has filed an S-1 ahead of its IPO. An S-1 is the long document a company files before it sells shares to strangers. It is mostly about money and worry.
As Scientific American reports, this filing spends real space on two things. One is how the lab hopes to keep its technology under control. The other is how far it hopes to push what AI can do.
In this version of events, the company did not trust the document to work on faith. So it tested it. One group of readers got the real filing. A second group got a placebo. That is a fake treatment built to look like the real one.
The placebo matched the real filing in length, font and staples. It described a small regional bakery. It listed the bakery's risks with the same care. Those included flour prices, a temperamental oven and "a customer who wants to talk about sourdough."
The results came in on a Wednesday. Readers of the bakery filing reported almost the same level of dread about advanced AI as readers of the real one. The gap between the groups was small. The odds that it happened by chance are low. They are not zero. Nothing is zero.

"Nineteen people in the placebo group asked whether the croissants were aligned," Dr. Fenn said. "I do not have a good answer for them. I am not sure anyone does."
The trial had limits, and Dr. Fenn listed them slowly. The sample was small. Some readers skipped the risk section. One person read the whole thing on a train and missed his stop twice. She said none of this changes the main finding. People are now so used to worrying about AI that a muffin can set them off.
The filing also lays out plans to use AI to speed up science. Dr. Fenn called that section "fine." Readers in both groups rated it as about as exciting as a parking policy. Several said they would read it later. None of them did.
Professor Odile Marsh studies how people read documents they did not ask for. She was not involved in the work. "This is a landmark result," she said. "It shows that the warning label now does the frightening on its own. You no longer need the product."
Lawyers for the bakery have asked to be removed from the study. They say the shop has received three calls from underwriters since Wednesday. The bakery has no plans to go public. Dr. Fenn said that is what everyone says at first.


