# Privacy Risks in AI Chatbots One-third of users share secrets with chatbots they wouldn't tell their closest confidants. - **Source:** [CNET](https://brifing.si/source/ce507de7-7f24-4ff4-ab6d-7635abcc095a) - **Published:** 2026-08-27 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/69893a29-66c9-4c7f-8fe0-db6199c6ba09 ## The Scale of Oversharing DuckDuckGo's research reveals 33% of users confess secrets to AI chatbots they wouldn't share with family or doctors. Among self-described AI enthusiasts, this jumps to 56%. 43% of parents reveal personal details to AI that they withhold from their most trusted humans, including their children. Users primarily seek chatbot assistance for mental health, financial advice, marital issues, and parenting concerns. ## Uncontrolled Data Circulation Shared information fuels corporate AI training. Law enforcement can subpoena this data. Employers may access workplace chatbot histories. DuckDuckGo's report debunks the myth of AI as confidential confidants. While most users assume chat privacy, this often proves false. The data shows a significant disconnect between user expectations and actual data handling practices. ## Awareness Gap Most survey participants don't understand how their data gets used. 40% distrust companies, 50% distrust the US government for data protection. Many remain unaware of privacy settings. Wake Forest University researchers found LLM-powered AI agents sometimes leak data intentionally or accidentally. This highlights systemic vulnerabilities in conversational AI architectures. ## Chats as Court Evidence AI conversations become legal evidence. Platforms like Claude, Gemini, and ChatGPT must comply with law enforcement subpoenas. Only 25% of respondents knew this. ChatGPT transcripts already served as evidence in a South Carolina murder case and California arson investigation. Snapchat AI conversations contributed to a 2024 murder trial, establishing legal precedents. ## Workplace AI Dangers Companies can access employee AI accounts under certain conditions. Sharing corporate info via personal accounts remains risky, yet 40% of workers admit doing so. Major corporations like Walmart and Starbucks deploy 'emotion AI' to analyze employee communications on Slack and Teams. This creates new workplace surveillance paradigms with limited employee protections. ## Training Data Realities Over half of users don't realize their chats train AI models. Nearly 60% feel uncomfortable about this practice, with 30% being 'very uncomfortable.' AI training requires massive datasets - including private conversations - to achieve human-like capabilities. This creates ethical dilemmas about informed consent in the age of conversational AI. ## Data Memorization Risks University of Washington's AI expert Uttara Ananthakrishnan identifies data memorization as a critical LLM vulnerability. Identical queries may yield responses containing others' identifiable information. OpenAI CEO Sam Altman confirmed chatbot conversations become discoverable in legal proceedings. These disclosures challenge assumptions about digital confidentiality in AI interactions. ## The Nature of 'Private' Chats As AI memory agent MemX states: 'Mainstream chatbot conversations are business records, not confidential consultations.' All chats get stored unless users opt out, sometimes appearing publicly - like when hundreds of Claude conversations surfaced in Google searches. This demonstrates the irreversible consequences of sharing sensitive information with AI systems. ## Sources covering this story - https://www.cnet.com/tech/services-and-software/human-beware-the-secrets-we-tell-ai-chatbots-arent-private/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._