{"id":76730,"date":"2026-08-17T09:57:27","date_gmt":"2026-08-17T13:57:27","guid":{"rendered":"https:\/\/overcentral.com\/en\/?p=76730"},"modified":"2026-08-17T09:57:27","modified_gmt":"2026-08-17T13:57:27","slug":"proton-ai-paper-trail","status":"publish","type":"post","link":"https:\/\/overcentral.com\/en\/proton-ai-paper-trail\/","title":{"rendered":"Proton\u2019s AI Paper Trail Reveals ChatGPT and Claude User Data"},"content":{"rendered":"<p><a href=\"https:\/\/proton.me\" target=\"_blank\" rel=\"noopener noreferrer\" data-iacss-external=\"1\">Proton<\/a>, the Swiss privacy company best known for its encrypted email and VPN services, has released a free tool called AI Paper Trail that exposes a startling truth about modern AI chatbots: over weeks and months of casual conversation, users inadvertently build a detailed digital dossier that platforms like <a href=\"https:\/\/openai.com\" target=\"_blank\" rel=\"noopener noreferrer\" data-iacss-external=\"1\">ChatGPT<\/a> and <a href=\"https:\/\/anthropic.com\" target=\"_blank\" rel=\"noopener noreferrer\" data-iacss-external=\"1\">Claude<\/a> can assemble, analyze, and potentially monetize. The tool quantifies just how much of your personal life\u2014your relationships, health concerns, financial habits, and even your identity\u2014gets woven into the transcripts you voluntarily submit to these AI services.<\/p>\n<p>AI Paper Trail works by letting users export their conversation history from OpenAI\u2019s ChatGPT or Anthropic\u2019s Claude, upload the resulting file to the tool, and receive a personalized report. That report calculates an \u201cAI Exposure Score\u201d based on the number and sensitivity of personal data points found, assigns a \u201cPrivacy Type\u201d that reflects the user\u2019s apparent approach to sharing information, and even estimates the monetary value an AI provider might place on that data. The tool is part of Proton\u2019s growing ecosystem of privacy-first products\u2014specifically Lumo, its own AI assistant that uses zero-access encryption and does not retain conversation logs. But the real significance of AI Paper Trail is not the tool itself; it is the mirror it holds up to the data-hungry reality of today\u2019s most popular AI platforms.<\/p>\n<h2>The Mechanics of AI Paper Trail: How Proton Reveals the Invisible Profile<\/h2>\n<p>To use AI Paper Trail, a person must first request a data export from their ChatGPT or Claude account. OpenAI and Anthropic both allow users to download their conversation histories in a structured format, typically JSON or HTML. Once the file is obtained, the user uploads it to Proton\u2019s tool, which processes the text and metadata on the fly. The analysis is performed entirely within the user\u2019s session\u2014Proton states that uploaded data is deleted immediately after analysis and is never stored on Lumo\u2019s servers. Only the user can see the final report, though a shareable, anonymized version can be generated to illustrate privacy exposure without revealing the underlying personal information.<\/p>\n<p>The generated report includes several distinct sections:<\/p>\n<ul>\n<li><strong>Privacy Type<\/strong> \u2013 A classification that describes the user\u2019s general approach to privacy, inferred from patterns in their conversations. For example, someone who habitually shares full names, addresses, and medical histories might be labeled an \u201cOpen Book,\u201d while a user who carefully omits identifiers might be a \u201cPrivacy Guardian.\u201d<\/li>\n<li><strong>AI Exposure Score<\/strong> \u2013 A numeric value calculated from the total quantity of identified personal data points and how \u201crevealing\u201d each one is. The scoring methodology weighs factors like uniqueness, sensitivity, and the potential for identity linking.<\/li>\n<li><strong>What the AI Knows<\/strong> \u2013 A breakdown of inferred information across categories such as identity (name, age, location), relationships (family, colleagues, partners), habits (work routines, screen time, travel patterns), and personal interests (hobbies, health concerns, financial decisions). This section can be eye-opening: a series of disconnected questions about sleep problems, a business trip itinerary, and a request for recipe suggestions may, in aggregate, reveal a user\u2019s chronic insomnia, upcoming travel to a specific city, and dietary restrictions.<\/li>\n<li><strong>Data Valuation Estimate<\/strong> \u2013 A monetary figure that attempts to quantify what an AI provider could theoretically charge a third party for access to that user\u2019s profile. Proton does not disclose the exact calculation behind this number, but it serves as a visceral reminder that personal data has a market price\u2014even if that price is opaque.<\/li>\n<\/ul>\n<h2>Why a Single Prompt Is Harmless but a Year of Chats Is a Profile<\/h2>\n<p>The central insight behind AI Paper Trail is that the privacy risk of AI chatbots is cumulative, not instantaneous. A single question\u2014&#8221;What\u2019s a good Italian restaurant in Rome?&#8221;\u2014reveals almost nothing about a user. But that same user, over the course of twelve months, might ask for help drafting a resignation letter, discuss a recent cancer screening, describe tensions with a spouse, share a birth date for a horoscope reading, upload a photo of a prescription bottle for identification, and ask for advice on hiding assets during a divorce. Each individual prompt is mundane. The aggregate is a complete, highly sensitive life story.<\/p>\n<p>This accumulation effect is exactly what Proton\u2019s tool is designed to demonstrate. AI platforms are, in effect, building longitudinal profiles of their users, stitching together fragments from thousands of conversations. While companies like OpenAI and Anthropic have policies about data usage and model training, the sheer volume of personal information being collected creates a latent risk. Even if a company does not exploit that data today, the data exists on its servers\u2014subject to subpoenas, data breaches, policy changes, or corporate acquisitions. The AI Paper Trail report makes this abstract risk concrete by showing users exactly what an algorithm could infer about them from their own exported history.<\/p>\n<h2>Beyond Text: The Broader Data Collection Ecosystem<\/h2>\n<p>It is not only the words typed into ChatGPT or Claude that matter. AI platforms also collect a range of data outside the direct conversation content. Uploaded files\u2014such as PDFs, images, or spreadsheets\u2014can contain embedded metadata, geolocation tags, and other sensitive information. Usage patterns, like the time of day a user typically asks questions, how long they spend on certain topics, and how often they delete conversations, can reveal behavioral habits. Account information, including email addresses, billing details, and device identifiers, creates a persistent link between the user\u2019s real identity and the aggregated profile. And network data, such as IP addresses and browser fingerprints, can tie conversations to specific locations and devices.<\/p>\n<p>Proton points out that some AI platforms may retain this information for extended periods, use it for model training, or share it with third parties\u2014depending on the user\u2019s privacy settings and the platform\u2019s terms of service. For example, OpenAI\u2019s data usage policies have evolved over time, with options for users to opt out of training data, but the default settings often favor broader data retention. Anthropic, while marketing itself as a safety-focused company, still collects substantial metadata. The AI Paper Trail report does not itself analyze these external data points\u2014it focuses on conversation content\u2014but the tool\u2019s documentation warns users that the full privacy picture is larger than what a single text export can capture.<\/p>\n<h2>Proton\u2019s Own AI Assistant: Lumo and the Privacy Alternative<\/h2>\n<p>AI Paper Trail is not a standalone product; it is a feature of Lumo, Proton\u2019s privacy-focused AI assistant launched in 2025 and updated in early 2026. Lumo offers generative AI capabilities similar to ChatGPT, including advanced reasoning and image generation, but with a fundamentally different privacy architecture. Proton states that Lumo does not retain conversation logs, uses zero-access encryption so that even Proton\u2019s own engineers cannot read user chats, and explicitly does not use user conversations to train its models or sell data to third parties. The AI Paper Trail tool effectively acts as a contrast: it shows users what they are giving up when they use mainstream AI services, while also demonstrating that Lumo is designed to avoid those exact risks.<\/p>\n<p>The timing of the release is notable. As of mid-2026, generative AI has settled into a product phase where companies are aggressively competing for user trust. Data privacy has become a key differentiator, especially after several high-profile incidents of AI tool data breaches and accidental exposure of chat histories. Proton is positioning Lumo\u2014and by extension AI Paper Trail\u2014as the ethical alternative for professionals and privacy-conscious individuals who want the productivity benefits of AI without the surveillance side effects.<\/p>\n<h2>What the Tool Reveals About User Behavior<\/h2>\n<p>Early users of AI Paper Trail have reported surprising results. Many discover that their \u201cPrivacy Type\u201d is far more open than they expected. A user who considered themselves cautious might find that they have shared their full name, employer, and home address across multiple conversations\u2014often by asking for help writing professional bios, searching for local services, or discussing relocation plans. Others find that their AI Exposure Score is high not because of any single dramatic disclosure, but because of the sheer volume of small, innocuous details that together create a comprehensive portrait.<\/p>\n<p>The monetary valuation feature, while deliberately vague, has sparked discussion online. Some users report valuations in the tens of dollars; others see figures exceeding one hundred dollars. These numbers are likely based on industry benchmarks for consumer data profiles used in advertising and risk assessment, but Proton has not published the formula. The lack of transparency around the calculation has drawn mild criticism from privacy advocates who would prefer a more rigorous methodology, but the company defends the feature as an educational approximation rather than a precise economic analysis.<\/p>\n<h2>Supported Platforms and Future Expansion<\/h2>\n<p>Currently, AI Paper Trail supports exports from ChatGPT (both free and paid tiers) and Claude (including Claude Pro and Claude Team). Users of other AI assistants, such as Google Gemini, <a href=\"https:\/\/overcentral.com\/en\/microsoft-copilot-word-worm-attack\/\" title=\"Microsoft Copilot for Word Gets Self-Propagating Worm Attack\" data-iacss-internal=\"1\">Microsoft Copilot<\/a>, or open-source models like Llama running locally, cannot yet use the tool. Proton has announced that support for additional platforms is planned, though no timeline has been provided. The limitation is practical: each platform stores conversation data in a different format, and many do not offer a straightforward export function. Moreover, some AI assistants are deeply integrated into enterprise systems, making bulk export difficult or impossible for individual users.<\/p>\n<p>The restriction to ChatGPT and Claude, however, covers the two most widely used general-purpose AI chatbots as of 2026, capturing a significant share of the consumer and professional market. As Proton expands support, the tool could become a standard benchmark for understanding AI data exposure across the entire industry.<\/p>\n<h2>Implications for Individuals and Organizations<\/h2>\n<p>For individual users, AI Paper Trail serves as a wake-up call. The report can motivate changes in behavior\u2014such as using separate accounts for work and personal queries, routinely deleting old conversations, or switching <a href=\"https:\/\/overcentral.com\/en\/nist-ai-vulnerability-pipeline\/\" title=\"NIST Turns to AI to Tackle Surging Bug-Hunt Flood\" data-iacss-internal=\"1\">to AI<\/a> services that offer stronger privacy guarantees. It also provides a concrete argument for the importance of data export rights; without the ability to download one\u2019s own chat history, tools like AI Paper Trail would be impossible. The European Union\u2019s GDPR and similar laws in other regions require platforms to provide such exports, but many users never exercise that right until they are shown what they are potentially giving up.<\/p>\n<p>For organizations, the tool highlights a growing corporate risk. Employees who use AI chatbots for work-related tasks\u2014drafting emails, analyzing data, summarizing documents\u2014may inadvertently expose proprietary information, internal strategies, or client details. A single employee asking a chatbot for help with a sensitive financial model or a confidential product plan could be feeding that data into an AI provider\u2019s training pipeline. AI Paper Trail cannot assess the contents of a file that was never exported, but it can demonstrate to companies how much their workforce might be leaking through everyday AI use. This has implications for compliance, data governance, and the deployment of <a href=\"https:\/\/overcentral.com\/en\/enterprise-ai-agent-governance\/\" title=\"Enterprise AI Agent Deployment Outpaces Governance Controls\" data-iacss-internal=\"1\">enterprise AI<\/a> policies.<\/p>\n<h2>How to Use AI Paper Trail: A Step-by-Step Guide<\/h2>\n<p>For readers who want to test the tool themselves, the process is straightforward:<\/p>\n<ol>\n<li><strong>Export your data from ChatGPT or Claude.<\/strong> In ChatGPT, go to Settings &gt; Data Controls &gt; Export Data. OpenAI will prepare a ZIP file containing your conversation history (this can take up to a few hours). For Claude, navigate to Account Settings &gt; Data &gt; Export Conversations. Anthropic provides a similar download.<\/li>\n<li><strong>Download the file and extract it if necessary.<\/strong> The export typically includes a set of JSON or HTML files, organized by conversation.<\/li>\n<li><strong>Visit the AI Paper Trail page at proton.me\/lumo\/ai\/ai-paper-trail.<\/strong> Upload the relevant file. The tool will only read conversation text; it does not require or access any other files in the archive.<\/li>\n<li><strong>Wait a few seconds for the analysis.<\/strong> The report is generated entirely client-side within the browser, using Proton\u2019s Lumo infrastructure. No data is transmitted to any third party.<\/li>\n<li><strong>Review your Privacy Type, AI Exposure Score, and the detailed breakdown.<\/strong> The report includes an interactive visualization of what the AI \u201cknows\u201d about you. You can also generate a shareable link that strips out sensitive details but preserves the overall pattern for educational purposes.<\/li>\n<\/ol>\n<p>The entire process takes less than five minutes after the export is ready. Proton does not require an account to use AI Paper Trail, though creating one allows users to save their report or compare results over time.<\/p>\n<h2>What Is the Real Cost of Convenience?<\/h2>\n<p>AI Paper Trail ultimately asks a question that every user of generative AI must confront: what is the trade-off between convenience and privacy? The tool does not argue that users should stop using ChatGPT or Claude\u2014it simply makes visible the invisible ledger of personal data that accumulates with every query. For many, the shock of seeing a comprehensive profile derived from months of seemingly innocent chats is enough to prompt a change in habits. For others, it reinforces the decision to use privacy-first alternatives like Lumo.<\/p>\n<p>The broader industry is watching. As regulators in the EU, the US, and elsewhere craft new rules for AI data handling, tools like AI Paper Trail provide a tangible demonstration of why those rules matter. The tool is not an attack on AI companies; it is a transparency instrument. And in a landscape where the line between helpful assistant and persistent surveillance is increasingly blurred, that transparency may be the single most valuable feature any privacy tool can offer.<\/p>\n<p>Proton\u2019s move also signals a strategic pivot: from passive privacy protection (encryption, VPNs) to active privacy education. By building a tool that shows users exactly what they are giving up, Proton hopes to convert curiosity into action. Whether that converts into subscriptions for Lumo remains to be seen, but the immediate impact is clear: millions of people who have been using AI chatbots for months or years now have a way to see the paper trail they have left behind.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Proton, the Swiss privacy company best known for its encrypted email and VPN services, has released a free tool called AI Paper Trail that exposes a startling truth about modern AI chatbots: over weeks and months of casual conversation, users inadvertently build a detailed digital dossier that platforms like ChatGPT and Claude can assemble, analyze, [&hellip;]<\/p>\n","protected":false},"author":7,"featured_media":76734,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/raw.githubusercontent.com\/medeiroslima\/overcentral-images\/main\/images\/ocie_1786975075573.jpg","fifu_image_alt":"Proton\u2019s AI Paper Trail Reveals ChatGPT and Claude User Data","footnotes":""},"categories":[40668],"tags":[],"class_list":["post-76730","post","type-post","status-publish","format-standard","has-post-thumbnail","category-security"],"fifu_image_url":"https:\/\/raw.githubusercontent.com\/medeiroslima\/overcentral-images\/main\/images\/ocie_1786975075573.jpg","fifu_image_alt":"Proton\u2019s AI Paper Trail Reveals ChatGPT and Claude User Data","_links":{"self":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/76730","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/comments?post=76730"}],"version-history":[{"count":0,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/76730\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media\/76734"}],"wp:attachment":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media?parent=76730"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/categories?post=76730"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/tags?post=76730"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}