ChatGPT Can Reveal Personal Information From Real People, Google Researchers Show.

Artificial intelligence-driven chatbots have often been in the spotlight during this past year: whether it was psychological support for veterans, emergency services, or online assistance, they have certainly been in the news.

ChatGPT, the renowned chatbot from the company OpenAI, was the subject of an investigation by a team of Google researchers. This analysis revealed a troubling capability of the model in question: having been "trained" through huge amounts of data collected from the Internet, ChatGPT is able to disclose private information of real people (we are talking about email addresses, legal names and phone numbers).

The team invested $200 in queries provided to the chatbot, resulting in more than 10,000 training examples stored within the service. With a larger budget, attackers would be able to extract much more data.