Study finds leading AI companies train LLMs on chatbot conversations

A new study examining the privacy policies of frontier artificial intelligence (AI) developers has revealed that six leading US companies use user conversations to train their large language models (LLMs). The study, led by Jennifer King from Stanford University's Institute for Human-Centered Artificial Intelligence (HAI), highlighted significant privacy risks as some companies, such as Anthropic, now use user inputs by default for AI training, unless the consumer actively opts out.

The research identified several concerns, including lengthy data retention periods and a general lack of transparency about how personal data, including information shared in dialogues, is used for training. King advises that users should be extremely cautious about the information they share with chatbots like ChatGPT and Gemini and should always opt out of data-for-training policies where the option exists.

Training Announcement: Freevacy offers a range of independently recognised professional AI governance qualifications and AI Literacy short courses that enable specialist teams to implement robust oversight, benchmark AI governance maturity, and establish a responsible-by-design approach across the entire AI lifecycle. Find out more.