On May 13, 2026, MIT Technology Review revealed that Google’s AI chatbot Gemini and other generative AI tools are exposing users’ real phone numbers, causing unwanted calls and privacy violations. A Reddit user reported his number was leaked for about a month, attracting calls from strangers seeking services like lawyers and locksmiths. Experts attribute this issue to personally identifiable information embedded in training data, warning that no simple fix currently exists to prevent such leaks.
The problem drew attention after a Reddit user posted in May 2026 about being overwhelmed by calls from strangers misdirected by Google’s AI. The user, whose account could not be independently verified, said his phone was flooded for about a month with calls requesting services such as lawyers and locksmiths. Comparable cases were reported by a software developer in Israel and a University of Washington PhD candidate, whose colleague’s phone number was leaked by Gemini during testing.
AI researchers and privacy experts have long cautioned about generative AI’s potential to expose personal data. A 2025 MIT Technology Review article noted that major AI training datasets include millions of personally identifiable information examples, which likely explains how chatbots like Gemini reveal real phone numbers. Despite this, the precise mechanism causing these leaks remains unclear. These incidents highlight broader privacy concerns linked to AI, with no straightforward solution available to prevent such exposures.
DeleteMe, a privacy-focused company that removes personal data from the internet, reported a 400% rise in customer queries related to generative AI over the past seven months. Rob Shavell, DeleteMe’s cofounder and CEO, said these queries—numbering a few thousand—specifically mention tools like ChatGPT, Gemini, and Claude. ChatGPT accounted for 55% of concerns, followed by Gemini at 20% and Claude at 15%, reflecting increasing user anxiety about AI surfacing sensitive data without consent.
The opaque nature of AI training data complicates efforts to address these leaks. Experts suggest that personally identifiable information, including phone numbers, may have been scraped from public or leaked datasets used to train models like Gemini. Without access to these datasets, identifying the source of leaks or creating effective safeguards is challenging. These incidents underscore the urgent need for stricter data governance in AI development to prevent further privacy breaches.
Individuals affected by these leaks face serious disruptions such as harassment and unwanted solicitations. The Reddit user described receiving calls from strangers seeking unrelated services, while the Israeli software developer was contacted via WhatsApp after Gemini gave incorrect customer service instructions containing his number. Similarly, a University of Washington PhD candidate’s colleague had their number exposed during casual interaction with the chatbot, illustrating the unpredictable nature of the problem.
Privacy advocates argue these incidents reveal a broader failure in AI regulation. Although companies like Google have implemented filters to block sensitive data, the leaks indicate these measures are insufficient. The lack of recourse for affected users worsens the issue, as there is no clear process to remove leaked information from AI models once exposed. This regulatory gap leaves users vulnerable to ongoing privacy violations without adequate protection or remedy.
Growing privacy concerns around AI coincide with increased scrutiny of tech companies’ handling of user data. In 2026, MIT Technology Review published multiple reports on AI’s role in surveillance and data privacy, including a January article on AI’s retention of personal information. These reports emphasize the urgency of addressing AI’s privacy risks, especially as generative tools become more integrated into everyday life and business, underscoring the need for stronger safeguards and regulatory oversight.