ChatGPT and Gemini are showing real personal phone numbers in some responses, a flaw that has already led to calls and messages to people unrelated to the services they supposedly offered. The problem points to the presence of identifiable data in the material with which these language models were trained.
The paradox is that tools presented as assistants capable of organizing and summarizing information are disseminating erroneous private data as if they were valid contacts. That leap between automation and personal exposure has multiplied user queries trying to erase their digital footprint.
DeleteMe registered a 400% increase in queries related to generative AI
The company DeleteMe has detected a 400% increase in queries linked to generative artificial intelligence assistants. Its co-founder and CEO explained to MIT Technology Review that many of these requests now expressly mention ChatGPT, Claude, Gemini, and other similar tools.
This change is not just a response to an abstract concern. The new cases describe how systems provide real phone numbers when answering questions about customer service, professional services, or the identification of specific individuals.
Unrelated users ended up receiving calls for services they do not provide
A Reddit user recounted that they began receiving calls from people asking for lawyer, product designer, and locksmith services, even though they do not work in any of those professions. In another case, a software developer's number appeared in Gemini's responses as a supposed customer service contact.
From there, their WhatsApp became saturated with external requests. The error was not just an imprecise response, but the assignment of a personal phone number to a commercial or support function that did not correspond to it.
A postgraduate student at the University of Washington also managed to get Gemini to provide her with a classmate's phone number. The episode reinforces concerns about the ease with which some models can return personal data when certain questions are asked.
AI hallucinations turn false data into a real contact channel
The problem is aggravated when the failures inherent in artificial intelligence validate incorrect information. If a model mixes a name, a profession, or a company with a real number, the response can circulate as if it were reliable even though the link is false.
There appears a second derivative. The data invented by the machine is no less harmful because the phone number actually exists, as it turns a specific person into the recipient of calls, messages, or queries that do not correspond to them.
The chief executive and co-founder of DeleteMe told MIT Technology Review that new user requests already make specific reference to ChatGPT, Claude, Gemini, and other generative AI tools.