People who use AI chatbots are confiding private matters to them, even as they doubt the companies running those tools. A Proton survey of 4,014 chatbot users found that 66% have discussed a sensitive topic with an AI assistant, according to the company's write-up of the research published today. The respondents came from the US, UK, France and Germany.
Trust in the companies behind those chatbots is thin. Only 11% to 20% of users reported high trust in AI companies with their private information, while 42% expressed little or no trust. Distrust ran highest in France, where 47.7% reported low trust, followed by the UK at 41.6%, Germany at 40.5% and the US at 38.8%.
Read more: Proton Drive Introduces Secure Photo Backup for iOS Users
The survey also asked what would change users' behavior. Just under half, 44%, said a guarantee that their chats would never train AI models would make them readier to disclose sensitive details. That figure was highest in the US at 47.7%, then the UK at 44.3%, Germany at 43.9% and France at 40.7%. More than 70% said they were interested in a chatbot built with privacy in mind, a share that reached 82% in the UK and 71.1% in Germany.
Sharing rates vary by country. In the US, 68.4% of users had raised one or more sensitive subjects with a chatbot, the highest share of the four countries, against 60.3% in France. Concerns about how chatbots use personal data are also widespread: 65.3% of US respondents worried about profiling for advertising, and 57.6% worried about their conversations being used to train models. In France, users reported being nearly as willing to ask an AI about money matters as to ask friends or family, a gap of 2 percentage points.
Proton says the survey was run unbranded, with participants not told which company commissioned it, to keep the sponsor's reputation from shaping answers. It was an online questionnaire of roughly 12 questions about chatbot usage. The company cautions that the results reflect what people say about themselves rather than their actual conversations, and that they do not represent non-users or countries outside the four surveyed.
Eamonn Maguire, director of AI engineering at Proton, framed the findings around how much context chatbots collect. "AI is learning far more about us than a search engine ever could. Search knows what you looked for. AI gets the hesitation, the follow-up questions, and the personal context between the lines," he said.
Proton has also published related work recently, including a piece asking whether Meta's smart glasses could help train robotics AI.
What the survey does not offer is a look at behavior outside self-reporting, or any data on people who avoid chatbots altogether.
