The information you type into AI chatbots may feel like a private conversation, but new research suggests those exchanges are far less confidential than users assume. According to survey findings from DuckDuckGo Research titled “The AI Privacy Problem,” roughly a third of people share secrets with chatbots that they withhold from friends, family, and medical professionals.
That figure climbs to 56% among self-described AI enthusiasts. The research also found that 43% of parents disclose things to AI tools that they don’t tell the most trusted people in their lives, including their own children.
The survey didn’t detail exactly what kind of personal information users reveal, but prior reporting has shown people frequently turn to chatbots for help with physical and mental health concerns, financial guidance, parenting questions, and marital issues.
Why Sharing With AI Carries Privacy Risks
The core issue isn’t seeking help from an AI tool. It’s losing control over what happens to the information after it’s typed into a chat. Companies can use private details to train their AI systems, law enforcement can obtain records through subpoenas, and workers who use chatbots on the job may find their chat history accessible to employers.
The DuckDuckGo report argued that AI companies promote chatbots as trusted confidants, describing that framing as a false narrative. “These messages all rest on one assumption: That these chats are private. In most cases, that just isn’t true,” researchers wrote.
Many survey respondents didn’t realize how their data could be used or accessed, or that settings exist to prevent conversations from being recorded and stored. When asked about trust, nearly 40% said they had no confidence in companies to protect their data, and close to 50% had no trust in the U.S. government to do so.
How Large Language Models Handle Your Data
Targeted advertising is already familiar territory. Searching for vacuums or Mediterranean cruises tends to trigger related ads, and companies routinely buy, sell, and share social media data to shape marketing. But AI data-sharing extends beyond product pitches. Sensitive information handed to a chatbot can be revealed or exposed.
A researcher at Wake Forest University found that AI agents, which use large language model data to perform autonomous tasks, can sometimes leak information either maliciously or unintentionally. Conversations don’t simply vanish. Unless a user opts out, the system records and saves them for future use, including additional model training. Last year, hundreds of chats from Anthropic’s chatbot Claude appeared in Google search results.
A central concern with large language models is data memorization, said Uttara Ananthakrishnan, assistant professor at the University of Washington’s Foster School of Business. “When someone asks the same question as someone else, it is possible that the LLM gives them exactly the same answer it gave someone else,” she said. “The LLM might not know that the answer contains information that someone can use to identify someone else.”
Law enforcement can also access chat data through legal requests, meaning material shared in a supposedly private exchange may end up in the hands of authorities.
