AI chatbots have become part of everyday life, and users share large amounts of information with them while asking questions and talking about work and personal life. A new tool from the company Proton sheds light on just how much these services can infer about their users.
اضافة اعلان
The Swiss company launched a free tool called "AI Paper Trail," aimed at showing what AI chatbots might know about a user.
The tool works by analyzing conversation history that can be exported from ChatGPT (OpenAI) and Claude (Anthropic).
To use the tool with ChatGPT, users export their data from account settings, receive a compressed file containing their data, then upload it to AI Paper Trail for analysis.
Proton says the data shared with its Lumo AI platform for analysis is deleted once the process is complete and isn't stored on its servers, and that the resulting report is only made available to the user themselves, according to what was published on Al Jazeera Net.
The tool's significance lies in the fact that it doesn't just display information the user explicitly typed, but also reveals the picture an AI system can build about them through the accumulation of conversations.
According to the outlet's test, ChatGPT was able to determine the user's name and profession, understand how they use AI in their professional and personal life, as well as their geographic location, hobbies, relationships, and some of the devices they use.
The test shows that a large volume of conversations over several months allows the chatbot to build a very detailed profile of the user.
A piece of information that seems trivial in a single conversation, according to the outlet, can become far more revealing when combined with dozens or hundreds of other pieces of information, illustrating that users may be revealing more about themselves to AI than they realize.
The AI Paper Trail tool also offers tips for reducing risk, including sharing only the minimum necessary information, avoiding disclosing too many details about family and location, reviewing privacy settings, disabling chat history or the use of data for model training where that option is available, and periodically deleting conversations and data.
Users can also turn to temporary chats when discussing topics they don't want a record kept of.
That said, an AI system knowing a lot about a user isn't always a drawback, since the better a chatbot understands a user's interests, context, and history, the more personalized and accurate its answers can become.
There is, therefore, a trade-off between privacy on one hand, and a smarter, more tailored experience on the other.
Proton is using this initiative to promote its Lumo service, which it says is privacy-focused and doesn't retain conversation logs or use them to train AI models, though other privacy-focused alternatives exist as well. Proton's tool reveals that a chat history with an AI system can turn into a comprehensive, detailed picture of a user's personality and life. As such, dealing with AI chatbots requires awareness of what we share with them, regularly reviewing privacy settings, and striking a balance between benefiting from personalization and protecting personal information.
Resource: Al-Ghad.
