ChatGPT Keeps Telling People 404 Media Can Solve Their Tech Support Problems

The outlet says a Russian man emailed its founder after ChatGPT suggested the site, and the chatbot wrote the English message he sent. When he asked whether he could email Meta, the bot said that was probably not possible, but eventually told him he should email the founder instead. Other senders include an Italian painter who said he was losing professional Meta accounts despite receiving a contradictory reinstatement email, and a Verizon customer who wanted an independent investigation into why their cell phone service wasn't working. Big platforms offer little human support, so people bring ordinary account problems to chatbots, which then route them to a small newsroom. The outlet has received dozens of AI-generated tips, including evidence dossiers and press releases, from people locked out of Instagram, Meta, or experiencing Verizon outages. New York Times reporter Kashmir Hill has reported a similar pattern of chatbots directing users to contact her. For anyone building an AI company, this is a product risk. A model that names specific outside contacts can push unsolicited volume onto third parties at scale, and it can invent escalation paths that do not exist. Teams should test how their models handle referrals to real people or outlets, limit the drafting of outbound messages to third parties, and decide what a model should say when no human support channel exists.