AI hallucinations are making entitled customers even worse

The Verge’s Gaby Del Valle reports frontline workers navigating guests who contradict allergy warnings, invent wine bottles, and dismiss sommeliers because a chatbot said otherwise. A New York server describes shellfish-allergy diners ordering fish broth after ChatGPT said it was fine. River rangers get AI itineraries that schedule upstream rows. An after-school teacher watches parents defend a ChatGPT sleep schedule against professional judgment.

Agentic booking and reservation emails are starting to show up too — requests that may never have been typed by a human. Meta’s Muse launch, the piece notes, is likely to amplify the mismatch between what an agent asks for and what a venue can actually do.

This is the other face of useful AI: tools that help people decide, used as weapons against the people who still have to deliver reality. We’re not anti-assistant. Asking a model for a wine pairing or a campsite plan can be fine. Treating that output as higher authority than a map, a menu, or a clinician is how you get entitlement with a citation.

The allergy cases are the bright line. Hallucinations aren’t a parlor trick when they override staff who are trying to keep someone out of anaphylaxis. Softening the blow with “maybe it’s showing you a French wine we don’t carry” is workers doing emotional labor for a model’s fiction.

I’m watching agentic commerce especially. If restaurants and parks start fielding bots that invent AV equipment and nonexistent campsites, the cost lands on low-wage and public-service labor first. Useful AI should make coordination easier — not train customers to argue with the only person who can see the kitchen.

The reporting sits alongside a wider wave of consumer agents — Muse, ChatGPT shopping, reservation bots — arriving before norms about when to trust a human over a model.