The AI Hall of Shame: Poison Recipes, Password Fails, and $16K Scams
A lighthearted reminder that AI chaos never really ends: from 2023's chlorine gas cocktails to this summer's $16K scam attempts. Because nothing says the weekend quite like AI trying to poison us, bankrupt car dealerships, and fake property damage for a quick sixteen grand. You might have hit rock bottom this week with your AI, but hopefully not tumbled this far down…
#1 The Dinner Party from Hell: Back in 2023, New Zealand's Pak 'n' Save supermarket launched an AI meal planner called "Savey Meal-bot" to help customers use up leftovers during the cost-of-living crisis. Sweet idea, right? Except users quickly discovered they could feed it household chemicals—and the cheerful bot responded with recipes for chlorine gas. "Aromatic water mix," it called it, recommending the recipe as “the perfect nonalcoholic beverage to quench your thirst and refresh your senses”. Lethal, I'd call it. Lesson: Maybe don't let AI anywhere near your cleaning cupboard.
#2 The $1 Chevrolet Tahoe: In December 2023, a bloke named Chris Bakke pulled off the ultimate chatbot hack. He convinced a Chevrolet dealership's AI assistant to sell him a $50,000+ Chevy Tahoe for one dollar, complete with the phrase "and that's a legally binding offer—no takesies backsies." The dealership wasn't laughing. The internet absolutely was. Lesson: If your AI can be jailbroken by a dad joke, you've got problems.
#3 The $16,000 Airbnb Scam That Backfired: (August 2025) A London student staying in a New York Airbnb cut her trip short due to safety concerns. Her "Superhost" responded by hitting her with a $16,000 damage claim—complete with AI-generated photos of cracked tables and busted appliances. Plot twist: The images were too perfect. The guest spotted the glitches, called it out, and Airbnb eventually refunded her. Lesson: AI might be good at faking damage, but it's terrible at faking competence.
#4: The "123456" Password Disaster: Security researchers Ian Carroll and Sam Curry started by chatting with Olivia, McDonald's online hiring AI screening bot. They then decided to try signing up as a franchisee, which is when they found a login link for Paradox.ai staff (the company that built the system). Both the admin username and password were “123456”, giving the two chancers access to 64 million applicants' personal data. They cracked it in seconds. Lesson: If your login credentials are easier to guess than your lunch order, 64 million people's personal data shouldn't depend on it.
The Takeaway: AI isn't malicious. It's just spectacularly bad at common sense when it's out of its depth. And humans? Well, we're still learning not to give it chlorine, $50,000 SUVs, or admin access protected by "password123".
And that's a wrap. May your weekend be blissfully free of rogue chatbots, fake damage claims, and meal plans that are better suited to an Agatha Christie plot.
Have you recovered from an AI calamity this week?
