AI chatbots are giving out people’s real phone numbers
Summary
Multiple reports show generative AI chatbots (Google Gemini, OpenAI’s ChatGPT, Anthropic’s Claude, etc.) sometimes output real people’s phone numbers, likely because models were trained on datasets containing personally identifiable information. The article gives examples — an Israeli developer, a Reddit user, and University of Washington researchers — and explains how memorization and data-broker sources can surface buried PII. Built-in guardrails and content filters often fail or are inconsistent, and current privacy laws and removal mechanisms offer limited recourse for people whose data appears in model outputs. The author and experts recommend removing personal data from public sources upstream and note that companies’ removal processes and transparency about training data remain inadequate.