AI Chatbots Are Giving Out People’s Real Phone Numbers

Why It Matters

Unfiltered personal data from AI models erodes user trust and exposes individuals to targeted harassment, while regulators lack clear authority over scraped public information used in training.

Key Takeaways

• Gemini and ChatGPT returned real phone numbers to user queries
• DeleteMe saw AI‑privacy complaints rise 400% in seven months
• Training data scraped from the web contains millions of PII records
• Current model guardrails often miss or bypass phone‑number filters
• No legal framework forces AI firms to delete scraped personal data

Pulse Analysis

The rapid adoption of large language models has outpaced the industry’s ability to safeguard personal information. Models are trained on petabytes of publicly available text, which inevitably include résumés, forum posts, and data‑broker listings that contain phone numbers and other identifiers. When a user asks a seemingly innocuous question, the model can retrieve and reproduce that exact data, as seen in recent Gemini and ChatGPT incidents that left individuals receiving unwanted calls and messages.

These leaks expose a regulatory blind spot. Existing statutes such as the California Consumer Privacy Act and the EU’s GDPR focus on data directly collected by companies, not on information scraped from the open web and later embedded in AI models. Data‑broker practices further complicate the picture, with dozens of brokers admitting to selling personal records to generative‑AI developers. While firms deploy content filters and privacy‑focused prompts, research shows that memorization effects enable models to reproduce verbatim PII, rendering technical safeguards insufficient on their own.

Industry and users must adopt a two‑pronged approach. Companies should invest in systematic data‑curation pipelines that identify and excise phone numbers and other sensitive fields before training, and they need transparent mechanisms for individuals to request removal of their data from model outputs. Meanwhile, consumers can reduce exposure by removing personal details from publicly searchable sites and leveraging data‑broker opt‑out tools. As AI continues to lower the barrier to accessing personal data, proactive privacy engineering and clearer legislative guidance will be essential to protect individuals from unwanted surveillance and harassment.