A recent analysis using Proton's AI Paper Trail tool has shown that uploading ChatGPT chat histories can reveal detailed personal information, such as financial details, health data, and psychological traits, even when users don't explicitly share such information. The tool, developed by privacy-focused engineers at Proton, uses Lumo — a large language model designed with privacy in mind — to analyze chat histories and generate user profiles. According to Proton, the tool deletes user data immediately after analysis and does not retain it. The study found that AI chatbots can infer personal details like age, income level, education, and political views based on patterns in how users interact with the systems. The way users write and speak in their prompts can also reveal socioeconomic status and psychological states, including stress and moments of vulnerability. This data can have monetary value, as companies may use it for targeted advertising or other commercial purposes. A 2026 data privacy ranking by Incogni assessed 13 AI platforms based on their privacy risks. While no platform was completely safe, some posed fewer risks than others. Meta AI, Gemini, and Copilot were among the worst performers, while Mistral's Vibe had fewer privacy concerns. However, Mistral's Vibe still uses publicly available data for training without fully disclosing where that data comes from. ChatGPT was ranked as the least invasive in terms of privacy, but recent investigations have revealed that humans are reading users’ prompts to help improve OpenAI’s models. These companies have not responded to questions about their privacy policies. A survey by DuckDuckGo found that more than a third of AI users have shared information with chatbots that they would not share with close friends, parents, or doctors. Over half of the surveyed users weren’t sure if their chatbot conversations were being used to train the AI systems. Even deleting chatbot conversations may not fully protect privacy, as data might still be stored in backup logs, server logs, or third-party processors. Some platforms may keep chat histories, uploaded files, account details, and usage patterns for long periods, even if users choose not to allow their conversations to be used for training. Experts recommend treating every chatbot interaction as potentially visible to others and using AI services that prioritize privacy, such as Proton's Lumo, xPrivo, and Internxt, which claim not to train on user data or store it on remote servers. If users continue to use major AI platforms, they are advised to use browsers instead of mobile apps, which often collect more device data and share it with third parties.