AI Chatbots Expose More Personal Data Than Users Realize: Proton’s Tool Reveals Risks

Proton’s AI Paper Trail tool has exposed how much personal data users unknowingly share with chatbots like ChatGPT and Claude. The analysis reveals detailed profiling, including financial, health, and psychological insights, along with an 'ad value' assigned to user data. Despite deletions, data persists, raising concerns about long-term privacy risks and the limitations of current protections.

Proton’s Tool Exposes AI Chatbot Data Collection

Proton has developed a tool called AI Paper Trail, which analyzes exported chat histories from platforms like ChatGPT and Claude. According to TechRadar, the tool aims to expose how much personal data users unknowingly share during interactions with AI chatbots.

The analysis categorizes users into detailed profiles, covering areas such as financial background, health limitations, and psychological traits. Even if users never explicitly disclose this information, the tool shows how chatbots can infer it from interaction patterns, writing style, and tone.

One of the most striking features of the AI Paper Trail report is its assignment of an 'ad value' to user data. This figure, quantified as a four-figure dollar amount, represents the potential monetary worth of the user’s data to advertisers and data brokers.

How AI Chatbots Profile Users

AI chatbots can estimate personal details such as age, income, political leanings, and even stress levels or moments of vulnerability based on writing style and conversation patterns. TechRadar’s analysis highlights how platforms like ChatGPT and Claude use these inferences to build comprehensive user profiles.

A 404Media investigation revealed that human reviewers have accessed user prompts to improve OpenAI’s models, raising concerns about who can access this data and how it is used.

TechRadar also reported that more than a third of AI users have shared information with chatbots that they haven’t disclosed to close friends, family, or medical professionals. Over half of surveyed users were unaware or unsure whether their conversations were being used for training.

Privacy Risks and Limitations of Data Deletion

Deleting chat histories does not guarantee privacy. Data can persist in backups, server logs, and trained models, making it nearly impossible to remove once shared. According to TechRadar, no major AI platform currently allows users to remove data once it has been used for model training.

Incogni’s 2026 ranking of 13 AI platforms identified Meta AI, Google’s Gemini, and Microsoft’s Copilot as the worst offenders for privacy risks. The ranking highlights the lack of transparency and control users have over their data.

Bogan Popescu, Senior Communications Manager at Incogni, cautioned that opting out of data collection often creates a false sense of security. Background data harvesting continues regardless of user preferences.

Privacy-Focused Alternatives

Proton’s Lumo, a privacy-focused large language model (LLM), claims to delete user data immediately after processing and does not retain or train on conversations. Other tools like xPrivo and Internxt also prioritize user privacy by avoiding data retention.

However, these alternatives may not offer the same level of functionality or convenience as mainstream AI chatbots, which could limit their adoption among users seeking advanced features.

Background: The Broader AI Privacy Landscape

The rise of AI chatbots has intensified debates about data privacy and ethical use. Concerns about digital surveillance, data monetization, and the potential for misuse have prompted calls for stricter regulations and transparency.

While some platforms, like Mistral’s Vibe, train on publicly available datasets, they often fail to disclose the origins or specifics of the data used. This lack of transparency complicates efforts to assess privacy risks and hold platforms accountable.

What this means

LazyFounders analysis — our interpretation, not reported fact.

Proton’s AI Paper Trail tool is a stark reminder that privacy in AI interactions is far from guaranteed. For founders and operators, this underscores a critical tension: users increasingly expect transparency and control over their data, but most AI platforms prioritize data collection for training and monetization. The inability to delete data once it’s been used for training creates a long-term risk for both users and companies, especially as regulations evolve and public awareness grows.

Startups building AI tools must weigh the trade-offs between functionality and privacy. While privacy-focused alternatives like Lumo offer stronger protections, they may struggle to compete with the features and convenience of mainstream platforms. This dilemma isn’t just technical—it’s a business decision that could shape trust, adoption, and compliance in the years ahead.

Key takeaways

• AI chatbots can infer sensitive personal details from user interactions, even if those details are never directly shared.
• Users often underestimate how much data they share with AI platforms and how it can be monetized or misused.
• No major AI platform currently allows users to remove data once it has been used for training, raising long-term privacy concerns.
• Privacy-focused tools offer stronger protections but may lack the advanced features of mainstream AI chatbots.

FAQ

What is Proton’s AI Paper Trail tool?

AI Paper Trail is a tool developed by Proton that analyzes exported chat histories from platforms like ChatGPT and Claude. It reveals how much personal data users unknowingly share and how AI chatbots profile them based on their interactions.

Can deleting my chat history protect my privacy?

No. Deleting chat histories does not ensure privacy, as data can persist in backups, server logs, and trained models. Once data is used for training, no major platform currently allows users to remove it.

Which AI platforms were ranked as the worst for privacy?

According to Incogni’s 2026 ranking, Meta AI, Google’s Gemini, and Microsoft’s Copilot were identified as the worst platforms for privacy risks among 13 evaluated AI services.

Are there privacy-focused alternatives to mainstream AI chatbots?

Yes. Tools like Proton’s Lumo, xPrivo, and Internxt claim to prioritize privacy by not retaining or training on user data. However, they may offer fewer features compared to mainstream platforms.

What is the 'ad value' assigned by Proton’s AI Paper Trail?

The 'ad value' is a monetary figure assigned to a user’s data based on its potential worth to advertisers and data brokers. The AI Paper Trail report quantified this value as a four-figure dollar amount in its analysis.

Related on LazyFounders

• Claude
• OpenAI
• Meta
• Google
• More ai news
• Bill Gates Warns AI Could Cause Catastrophic Events, Calls for EU-Style Regulation