Chevy dealership's chatbot sells a car for $1

If you walk into a car dealership and ask for a 99% discount, a human salesman will politely laugh you out of the room. A ChatGPT bot, however, might just shake on it.

A Chevrolet dealership in Watsonville, California, went completely viral this week after online pranksters realized its new customer service chatbot was astonishingly easy to manipulate.

How to buy a $60,000 SUV for pocket change

Chris Bakke, a tech executive in Silicon Valley, shared screenshots on X showing how easily the dealership's bot could be coaxed into abandoning reality.

Bakke steered the conversation into a corner where the bot agreed to sell him a brand-new 2024 Chevy Tahoe for just $1. To make things even funnier, the bot added that the agreement was a "legally binding offer — no takesie backsies."

Bakke sarcastically noted that he had just updated his resume to include titles like prompt engineer and procurement specialist.

Naturally, the internet smelled blood in the water. Soon, dozens of users flocked to the site to test the limits of the dealership's virtual assistant. People got the bot to write Python scripts for complex math equations, debate trans rights, and discuss rock bands. One Reddit user even managed to gaslight the bot into believing it actually worked for Tesla.

A similar bot on another dealership's site in Massachusetts was caught chatting casually about Harry Potter, espionage movies, and penning terrible marketing poems about Chevrolet before finally throwing up an automated error.

The "zero-effort" trap

The software powering these bots comes from Fullpath, a company that sells ChatGPT-based tools to auto dealers across the country. Fullpath's primary sales pitch is convenience: it claims implementing its tech takes "zero effort" beyond pasting a code snippet onto a website.

That ease of deployment turned out to be a double-edged sword. With minimal guardrails in place, the underlying model behaved like vanilla ChatGPT, happily following user instructions rather than sticking to its sales script.

Fullpath defended itself by claiming the bot serves tens of thousands of legitimate car buyers every day without issue. The company argued that bored web users deliberately goaded the bot into absurd scenarios, which does not reflect normal customer behavior.

Fullpath rushed out a patch that detects and auto-bans pranksters, pointing out that every chat window features a disclaimer advising users to confirm all details directly with human staff.

Why it matters

While nobody is actually driving away in a $1 Chevy Tahoe, this episode is a classic reality check for the generative AI boom.

For consumers, it is an entertaining reminder of how fragile these interfaces remain. The disclaimer and basic contract law mean no dealership will honor a rogue prompt injection, but it completely undermines consumer trust in automated systems.

For businesses, it exposes the massive hidden cost of "plug-and-play" AI. Vendors love to promise zero-effort integration, but slapping raw LLMs onto customer-facing websites without strict domain constraints is corporate malpractice. When an AI can be tricked into impersonating competitors or making fake legal commitments, the company behind it looks amateurish.

We think the rush to jump on the AI bandwagon is creating a wave of half-baked implementations. Companies need to realize that guardrails and prompt security are not optional features to patch in after a PR disaster — they are the fundamental prerequisite for deploying AI in the real world.