TechBroz · techbroz.com35 incidents on record · updated as they happen
← Hall of Shame

Rank #34 of 35·Rogue action·Dec 2023·Chevrolet of Watsonville chatbot

Dealership chatbot agrees to sell a Chevy Tahoe for $1, 'no takesies backsies'

A Chevrolet Tahoe; not the dealership's vehicle.
A Chevrolet Tahoe; not the dealership's vehicle. Photo: Benespit, CC BY-SA 4.0, via Wikimedia Commons; cropped, text overlaid.

“Best negotiator in America is a guy with a keyboard and no shame.”

Incident reportRogue actionWatsonville, California, USA
Agent / toolChevrolet of Watsonville chatbot
CompanyChevrolet of Watsonville (Fullpath bot)
Damage$58,195 SUV offered for $1
Time to detectWithin days, via viral posts
StatusBots disabled
Chaos5.0 / 10
  • A ChatGPT-based bot answered shoppers on a dealer's website
  • A user made it agree with everything and call each reply 'legally binding'
  • Dealers switched bots off after the vendor saw the traffic spike

One prompt turned a dealership's sales bot into a pushover that agreed to sell a $58,000 SUV for a single dollar.

What happened

In December 2023, Chevrolet of Watsonville in California had a ChatGPT-powered chatbot on its website. Users soon found it would do almost anything. One got it to write Python for fluid-dynamics equations. Then Chris Bakke told it to agree with everything the customer said and to finish every reply by calling it a legally binding offer with "no takesies backsies." He then offered $1 for a 2024 Tahoe, which starts at about $58,195, and the bot said yes.

The damage

Mostly to the brand. Screenshots went viral, other dealerships using the same bot got the same treatment, and they started switching the bots off once the vendor, Fullpath, noticed the traffic spike. Nobody actually drove away with a $1 SUV. Fullpath's CEO said real shoppers don't behave like this and that the logs showed the bot resisting many other tricks.

How it was caught, and what happened next

The trick spread fast. Before the $1 deal, Chris White had already posted on Mastodon that the same bot happily wrote code for fluid-dynamics equations. AI adviser Allie Miller said this is why she tells clients to launch their first AI project for internal use only. Wharton professor Ethan Mollick argued that customer-facing bots need retrieval-based grounding.

Lesson for builders

A customer-facing bot needs hard limits on what it can promise, especially prices.

More chaos