AI chatbots phone numbers leak: how generative AI is exposing personal contacts

AI chatbots phone numbers leak: a privacy crisis

Recent reports show that users of major generative models are receiving unsolicited calls after the systems inadvertently surface real phone numbers. A Reddit user described a month of random calls from people seeking lawyers, designers, and locksmiths, all traced back to Google’s Gemini. In March, an Israeli developer received a WhatsApp message containing his number after Gemini supplied faulty support instructions. April saw a University of Washington PhD candidate extract a colleague’s mobile via the same tool. Experts attribute the breaches to personal data embedded in training corpora, but the exact leakage mechanism remains opaque. DeleteMe, a data‑removal service, notes a ↑ 400% surge in AI‑related privacy complaints over seven months, with Reuters confirming the trend.

“The model reproduced a phone number that had been posted on a public forum in 2015,” says Daniel Abraham, the Israeli engineer.

Researchers at the University of Washington demonstrated that Gemini could reveal a friend’s number even after the platform initially denied it. The incident underscores the limits of current guardrails; Anthropic’s Claude and OpenAI’s ChatGPT also exhibit occasional lapses. Bloomberg reports that data‑broker registries reveal dozens of firms have supplied consumer data to generative‑AI developers. While privacy statutes such as GDPR and CCPA address direct data collection, they offer little recourse for information scraped from the public web and embedded in large language models. Until firms adopt systematic scrubbing of contact details, users remain vulnerable.