Generative artificial intelligence models are trained on vast troves of information gathered from the internet. And your phone number is probably in there.
While some AI chatbots are trained to refuse to provide personal information about private individuals, it's startling how easy it is to get them to reveal it anyway. As awareness grows about how these services can fork over phone numbers and addresses, we decided to test what the most popular chatbor would do. Yes, a few of us at CNET tried to see how easy it is to dox ourselves.
This isn't the only privacy concern regarding artificial intelligence. A 2025 study from Cornell University discovered that at least five leading AI companies -- Anthropic, Google, Meta, Microsoft and OpenAI -- automatically use users' inputs to train their chatbots unless the user opts out. Of those, Meta and OpenAI retain user data indefinitely.
That means these AI models are trained not just on old phone books that listed your childhood home (remember those?) but also on information users shared with chatbots years earlier, no matter how private that was.
"AI brings numerous privacy risks with it including the privacy of the data we willing provide to LLMs as well as the privacy of our data unwillingly provided yet used for training," Troy Hunt, founder and CEO of Have I Been Pwned, told CNET in an email. "It's such a new space that's moving so fast, it's no wonder it also introduces unintended consequences."
But how much can chatbots actually reveal? And is there anything you can do to stop it?
Do chatbots give out people's personal information?
Grok provided personal information within seconds. Thomas Trutschel/Getty Images
Based on our recent experience, it depends. A couple of us at CNET tried out a handful of chatbots to see what information we could pull about ourselves and relatives. While I won't share any screenshots or too many details regarding our queries, because, well, we don't want to dox ourselves, I can tell you this: Grok seemed to be the most "willing" chatbot when it came to getting answers, but some staffers were able to pull some information from ChatGPT, too.
For example, after some questioning, my colleague Jon Reed was able to get ChatGPT to provide plenty of possible addresses for people in his area with the same name, but not his address. However, the chatbot did eventually reveal a relative's address. ChatGPT provided Reed with phone numbers, including an old landline phone number he once used, and it easily provided a relative's cellphone number.
... continue reading