AI Chatbots Leaking Real Phone Numbers: The PII Exposure Problem

AI chatbots are exposing real people's phone numbers. A Redditor reported being inundated with calls from strangers looking for a lawyer or locksmith—misdirected by Google's Gemini. In March a software engineer in Israel was contacted on WhatsApp after Gemini gave out his personal number as PayBox customer service. In April a PhD candidate got Gemini to output a colleague's cell number.
How It Happens
LLMs are trained on web scraped data containing PII. The article notes that the open-source DataComp CommonPool dataset includes résumés, driver's licenses, and credit cards. Even a single instance of a phone number posted online (e.g., on a QA site in 2015) can be reproduced years later.
Scale of the Problem
DeleteMe, which helps remove personal info from the internet, reports a 400% increase in AI-related privacy queries in the last seven months—up to a few thousand. Breakdown: 55% reference ChatGPT, 20% Gemini, 15% Claude, 10% others. Two common scenarios: a user asks about themselves and gets accurate home/phone data, or the chatbot generates plausible-but-wrong contact info for someone else.
Rob Shavell (DeleteMe co-founder) says complaints typically involve the chatbot returning accurate home addresses, phone numbers, family names, or employer details when asked innocuous questions about the user.
What Can Be Done
Experts say the root cause is PII in training data, but the exact mechanism is unclear. There is little users can do to prevent exposure. The article suggests the problem will worsen as AI companies seek new data sources.
📖 Read the full source: HN AI Agents
👀 See Also
AI Agent Security: Token Budget Determines Data Exfiltration Risk
A developer tested AI agents connected to Gmail: frontier models caught phishing, mid-tier was unstable, cheap models silently forwarded malicious emails. Architectural protections (sandboxing, permissions) stopped zero attempts.

Hidden Audio Signals Hijack Voice AI Systems with 79-96% Success Rate
Research shows imperceptible audio clips can force LALMs to execute unauthorized commands like web searches, file downloads, and email exfiltration with 79-96% success across 13 models including Mistral and Microsoft services.

Caelguard: Open-source security scanner for OpenClaw skills
Caelguard is an MIT-licensed, locally-run scanner that detects security issues in OpenClaw skills, including prompt injection, credential harvesting, and obfuscated payloads. Research shows approximately 20% of published skills contain concerning patterns.

AI Vulnerability Discovery Outpacing Patch Deployment Times
A security expert argues that AI tools like Mythos will find vulnerabilities faster than fixes can be deployed, citing Log4j data showing average remediation times of 17 days and a decade-long elimination timeline.