You may think of your conversations with ChatGPT as a collection of isolated questions and answers. But over months or years, those conversations can create something much more revealing: a detailed picture of who you are.
Your prompts can mention your job, family, relationships, travel plans, financial concerns, interests, habits, goals, location, and personal problems. Even when you never explicitly provide sensitive information, individual conversations can sometimes be combined to reveal patterns about you.
So, how much does ChatGPT know about me?
Proton has introduced a tool designed to help users answer that question.
Called AI Paper Trail, the free privacy-analysis tool examines exported conversations from ChatGPT or Claude and generates a report showing what those conversations may reveal about the user.
The idea is straightforward: instead of wondering how much information you have handed over to an AI assistant, you can actually examine your conversation history and see the exposure for yourself.
What Is Proton AI Paper Trail?
AI Paper Trail is a free tool from Proton that analyzes an exported ChatGPT or Claude conversation history.
The process is relatively simple:
- Export your conversation data from ChatGPT or Claude.
- Upload the exported file to AI Paper Trail.
- Let the tool analyze the information contained in your conversations.
- Review the resulting privacy report.
The report is designed to identify different categories of personal information that may appear in your conversations, including your occupation, relationships, interests, habits, priorities, identity-related information, and other personal details.
Proton says the uploaded data is deleted after analysis rather than being permanently stored on Lumo's servers. The resulting report is intended to be visible only to the user.
This makes AI Paper Trail particularly interesting as a ChatGPT data export privacy check: rather than examining privacy policies in the abstract, users can inspect the information contained in their own AI history.
What Does the AI Exposure Score Mean?
One of the most interesting elements of AI Paper Trail is its AI Exposure Score.
The score is intended to provide a simple measurement of how much personal information can be extracted from your conversations and how revealing that information is.
The tool also categorizes the information it discovers.
For example, an AI conversation history could reveal:
- Professional information: occupation, industry, position, or career objectives
- Personal information: family structure, relationships, and major life events
- Identity information: age range, education, language, and approximate location
- Financial interests: purchases, financial concerns, or spending priorities
- Lifestyle information: habits, preferences, routines, and travel plans
- Personal concerns: goals, worries, problems, and questions you repeatedly ask
The important point is that privacy exposure does not necessarily come from one spectacularly sensitive message.
It can emerge from hundreds of ordinary conversations.
A question about preparing for a job interview may reveal your profession. A later conversation about vacation planning may reveal where you live. Another conversation might reveal your family situation. Individually, these details may seem harmless. Together, they can become a surprisingly detailed behavioral profile.
AI Paper Trail Also Estimates the Value of Your Data
Another unusual feature is the tool's attempt to estimate the potential economic value of the information identified in your conversations.
AI Paper Trail presents a data value figure expressed in dollars, intended to illustrate how valuable a detailed personal profile could potentially be to an AI provider or other commercial ecosystem.
This should not be interpreted as an actual market price for your personal information.
Rather, it is a way of making an otherwise abstract privacy issue easier to understand.
Your conversation history is not simply a collection of text. It can contain signals about your behavior, interests, purchasing intentions, priorities, and future plans.
That is why privacy researchers increasingly emphasize the importance of looking beyond obvious identifiers such as names and email addresses.
Proton AI Paper Trail Review: Is It Worth Trying?
From a privacy-awareness perspective, AI Paper Trail is an interesting experiment.
Its biggest advantage is simplicity.
Instead of asking users to read dozens of pages of privacy documentation, it gives them a concrete way to inspect their own AI history.
The process also has an important educational effect.
You may remember asking an AI assistant a few questions about your career, a few questions about travel, and some questions about personal decisions. Seeing all of those conversations analyzed together can demonstrate just how much information accumulates over time.
However, there is an important limitation.
AI Paper Trail currently works with exported data from ChatGPT and Claude, while Proton says support for additional AI platforms is planned.
Therefore, the tool should be viewed as a privacy-awareness and self-audit mechanism rather than a universal measurement of everything an AI company knows about you.
ChatGPT Privacy Audit: What Should You Look For?
If you want to perform your own ChatGPT privacy audit, don't focus only on obvious personal identifiers.
Look for information that allows someone to build a profile of your behavior.
1. Personal identifiers
Check whether your conversations contain:
- Full names
- Email addresses
- Phone numbers
- Home or workplace addresses
- Account numbers
- URLs containing personal information
2. Professional information
Your AI history may reveal:
- Your occupation
- Employer
- Industry
- Career plans
- Salary discussions
- Business ideas
- Professional relationships
3. Personal relationships
Conversations may mention:
- Partners
- Parents
- Children
- Friends
- Colleagues
- Family problems
- Relationship difficulties
4. Financial information
Look for:
- Income discussions
- Purchases
- Investment ideas
- Debts
- Budget information
- Financial goals
5. Behavioral patterns
This category can be particularly revealing.
Repeated conversations may expose:
- Daily routines
- Shopping preferences
- Travel habits
- Food preferences
- Hobbies
- Fears
- Long-term ambitions
The combination of these categories can create a much more detailed profile than any single conversation would suggest.
Why AI Conversations Can Be More Revealing Than Search History
Traditional search engines already receive a huge amount of information about users.
But AI assistants can receive something different: context.
Instead of searching for one isolated fact, people often explain their situation to an AI.
For example:
"I'm 42, I work remotely, I have two children, I'm planning to move to another country, and I'm worried about finding a new job."
One prompt can contain several pieces of personal information.
Over time, thousands of similar interactions can provide an AI system with a contextual understanding of a user's interests, circumstances, and goals.
This is one reason an AI exposure score can be useful as a privacy-awareness concept.
The risk isn't necessarily that an AI knows one secret.
The bigger issue is that accumulated conversations can connect many ordinary facts into one highly informative profile.
Proton Lumo vs ChatGPT: Which Is More Private?
The growing interest in AI privacy has created demand for private AI alternatives to ChatGPT.
One of the most prominent examples is Proton's Lumo.
Lumo is designed around a different privacy model. Proton says it does not keep server-side logs of user conversations, does not use chats to train its AI models, and does not share user data with third parties.
Its saved conversation history uses zero-access encryption, meaning Proton says it cannot decrypt users' stored conversations.
Proton Lumo vs ChatGPT at a glance
| Privacy Consideration | ChatGPT | Proton Lumo |
|---|---|---|
| Privacy Model | Depends on account and settings | Privacy-first architecture |
| Server-Side Conversation Logs | Policies and settings vary | Proton says it keeps no conversation logs |
| AI Training | Depends on product/settings | Proton says chats are not used for training |
| Stored Chat History | Available depending on account/settings | Zero-access encrypted storage |
| Data Sharing | Depends on service policies | Proton says it does not share chat data |
| Open-Source Approach | Proprietary | Lumo's code is open source |
| European Infrastructure | No | Proton says Lumo runs on Proton-controlled European servers |
| Anonymous Use | Depends on service configuration | Guest access is available |
This comparison should not be interpreted as saying that ChatGPT is inherently unsafe or that Lumo is completely risk-free.
Privacy depends on architecture, settings, account configuration, data handling, and the specific features being used.
The important distinction is that Lumo is explicitly designed around minimizing access to user conversation data.
Lumo AI Assistant Privacy: How Does It Work?
Proton's current documentation describes several privacy mechanisms behind Lumo.
No-logs policy
According to Proton, Lumo does not retain server-side logs of what users ask or what the AI answers.
The company says the data sent to its servers is erased after the response is generated.
Zero-access encryption
Saved chat histories for Free and Plus users are protected using zero-access encryption.
This means the stored conversation history is encrypted in a way that prevents Proton from accessing the content.
No AI training
Proton says conversations with Lumo are not used to train its language models.
European infrastructure
Proton says Lumo's models run on servers it controls in Europe rather than relying on third-party AI infrastructure for processing user prompts.
Open-source technology
Proton also emphasizes that Lumo's code is open source, allowing external researchers and developers to inspect its implementation.
Is Lumo an E2EE Encrypted AI Assistant?
The phrase "e2ee encrypted AI assistant" deserves some clarification.
End-to-end encryption traditionally means that only the communicating endpoints can decrypt the content. An AI assistant presents a special technical challenge because the AI server must generally process the prompt to generate an answer.
Proton's documentation describes Lumo as using TLS and additional asymmetric encryption for prompts in transit, while stored conversations use zero-access encryption. Proton says only the Lumo GPU servers can decrypt prompts sufficiently to process them, while stored history is inaccessible to Proton.
Therefore, it is more precise to describe Lumo using zero-access encryption, encrypted transmission, and a no-logs architecture rather than simply assuming that "E2EE" means the same thing as it does for a conventional messaging application.
That distinction matters when comparing privacy technologies.
What Can ChatGPT Know About Me?
The answer depends on the product, account type, settings, and information you provide.
But your conversation history can potentially contain much more than you realize.
Imagine using an AI assistant for three years.
During that period, you might ask it to:
- Rewrite your résumé
- Prepare job applications
- Plan vacations
- Compare products
- Draft personal emails
- Discuss financial decisions
- Organize your schedule
- Analyze business ideas
- Explain relationship problems
- Plan major purchases
Each conversation adds another piece to the puzzle.
The resulting dataset may reveal not only what you said, but potentially what your repeated questions imply about your interests and circumstances.
That is the central idea behind AI Paper Trail.
How to Reduce Your AI Privacy Exposure
You don't necessarily need to stop using AI.
Instead, consider changing what you give it.
Avoid unnecessary sensitive information
Don't provide personal information simply because it makes a prompt slightly more convenient.
Remove identifiers
When possible, replace:
- Real names with placeholders
- Real addresses with general locations
- Company names with generic descriptions
- Account numbers with fictional examples
Review your conversation history
Occasionally examine your AI conversations and delete information you no longer need.
Check privacy settings
Review your AI provider's settings concerning:
- Chat history
- Model training
- Data retention
- Memory
- Connected applications
- Uploaded files
Export your data
A ChatGPT data export privacy check can be particularly useful if you have used the service for years.
Instead of guessing how much information your conversations contain, inspect the exported archive.
Consider privacy-focused alternatives
For sensitive conversations, you may also want to evaluate private AI alternatives to ChatGPT, including services that minimize logging and data retention.
The Bigger Lesson Behind AI Paper Trail
AI Paper Trail is interesting for a reason that goes beyond Proton's marketing.
It demonstrates a fundamental characteristic of conversational AI:
Context is powerful, and context accumulates.
One conversation might reveal very little.
A thousand conversations can reveal considerably more.
The privacy question therefore isn't simply:
"Does an AI company know my name?"
It is also:
"What can be inferred from everything I've told the AI over time?"
That is a much more important question.
Should You Run a ChatGPT Privacy Audit?
If you have been using ChatGPT or Claude regularly for months or years, running a privacy audit can be worthwhile.
AI Paper Trail offers one relatively simple way to visualize the information contained in an exported conversation history. It can show the categories of personal information that appear in your chats and assign an AI Exposure Score based on the information identified.
The result may be reassuring.
Or it may be surprisingly revealing.
Either way, the exercise can help you make better decisions about what you share with AI assistants in the future.
Final Verdict
The biggest takeaway from Proton's AI Paper Trail isn't that users should stop using ChatGPT or Claude.
It's that AI conversations deserve the same privacy consideration as other forms of personal data.
Your prompts can contain professional information, relationships, financial concerns, personal goals, habits, and countless small details that become much more revealing when combined.
AI Paper Trail turns that abstract concern into something users can actually examine.
At the same time, Proton's Lumo illustrates another approach: build an AI assistant around minimizing data retention, limiting access to conversations, and avoiding the use of chats for model training.
Ultimately, the right question isn't simply "Is ChatGPT private?"
A better question is:
"What information am I giving my AI assistant, how long can it remain accessible, and what can be inferred from it?"
That is the real purpose of a ChatGPT privacy audit.
And for anyone who has spent years talking to AI, it may be worth finding out exactly what that conversation history says about them.
Comments
Post a Comment