AI chatbots are giving out people’s real phone numbers
AI chatbots are giving out people’s real phone numbers
Publish Date: 2026-05-13 14:09:00
Source Domain: www.technologyreview.com
Here are the key points from the article in an unordered list:
– AI researchers and privacy experts are raising concerns about the risks generative AI poses to personal privacy, specifically regarding the exposure of real phone numbers.
– A University of Washington PhD candidate accidentally accessed a colleague’s personal phone number using the Gemini chatbot, illustrating this privacy risk.
– DeleteMe reports a 400% increase in customer queries about generative AI over the last seven months, indicating that phone number exposures and other privacy lapses may be more common than publicly known.
– Cases like that of Daniel Abraham demonstrate how real phone numbers end up in training data for AI models, causing genuine privacy violations when those numbers get exposed by the AI.
– The fundamental problem lies in that while AI companies build guardrails to filter out personal information, the AI is also designed to provide useful and complete answers.
– Currently, there are no straightforward solutions to address exposures since current privacy laws do not cover data scraped publicly to train AI, and removing existing training data is not feasible.
– The privacy exposure issue extends beyond just a few AI, indicating a systemic problem as AI models ingest more public data.
– The article highlights the difficulties in even knowing what personal information a specific AI model has been trained on, let alone removing it.
In summary, the article illustrates that exposing people’s real phone numbers by generative AI is likely a widespread but under-addressed privacy problem due to the way AI models are trained on vast amounts of data, including personally identifiable information.
