Study: AI services leak user data to advertisers

Hacker News is actively discussing a PDF document about how companies building AI services share their users' data with advertisers.

The document itself carries a playful name — "Prompt like a butterfly, sting like a tracker" — a riff on the famous line about a boxer's style. Behind the joke lies a serious claim: what users type into conversational AI services does not stay strictly between them and the model, but ends up within reach of advertising tracking machinery.

The full text of the study could not be retrieved, so the specifics of its methodology, the list of named companies, and the scale of the leaks are not reproduced here. What is known is that the piece drew a notable response from the community: at the time of discussion it had gathered 157 points and 39 comments.

The topic naturally resonates with long-standing worries about privacy in the age of generative AI: the more personal information users entrust to chatbots, the higher the cost of any leak of that information to third parties.

Source: Hacker News