News
3 min read
Swiss company Proton has released a free tool that allows users to check how much personal information ChatGPT and Claude have built up from their chats. The tool is called AI Paper Trail and is part of Lumo, Proton's own AI assistant. Users export their conversation history from ChatGPT or Claude, upload that file, and then receive a report detailing what the relevant model appears to know about them.
Proton was founded in Geneva in 2014 by scientists who met at CERN. The company is known for privacy-focused alternatives to widely used services, such as ProtonMail and ProtonVPN. With AI Paper Trail, it is now turning its attention to the information advantage that major AI providers hold over their users.
Eamonn Maguire, Director of Engineering at Proton, says the tool is intended to help people understand what they are sharing with AI systems. AI Paper Trail also has a clear commercial objective: the tool explicitly steers users towards Lumo as an alternative.
To use AI Paper Trail, a user first exports their conversation history from ChatGPT or Claude. That export file is then uploaded to the tool, which analyses the content and generates a personalised report.
The report sets out what personal information can be inferred from the conversations, such as profession, relationships, habits, interests and priorities. It also includes a score indicating how easily the user can be profiled based on their chat behaviour, and an estimate of the so-called data value, an indication of what that profile information could be worth commercially.
Proton states that uploaded conversation data is deleted immediately after analysis and is not retained on Lumo's servers. The generated report is visible only to the user themselves. Independent verification of that claim is not possible on the basis of the available source material.
For now, the tool works with exports from ChatGPT and Claude. Proton has indicated it will add support for other AI platforms in the future.
AI Paper Trail illustrates how Proton combines its privacy mission with its commercial strategy. The report generated by the tool explicitly points users towards Lumo as a switching option at the end. This makes the tool both an awareness instrument and a customer acquisition vehicle.
That need not undermine the substantive value of the analysis. The fact that large language models can distil personal information from conversations is not technically surprising, but most users do not actively reflect on it. A tool that makes this tangible using a person's own data has a different impact than an abstract warning in a privacy policy.
Proton has for years sought to be seen as a privacy-focused alternative to services offered by large American tech companies. That positioning has gained commercial relevance as European regulation, in particular the GDPR and the AI Act, places greater pressure on how companies handle personal data in AI systems.
AI Paper Trail fits into a broader movement in which European tech companies use transparency around data processing as a differentiating feature against American competitors. Proton has been doing this with its email and VPN services; Lumo is the extension of that strategy into the AI assistant market.
For founders and investors in the European AI sector, this confirms that privacy compliance is increasingly functioning as a sales argument, not merely a legal obligation. Companies that build on principles such as data minimisation and transparency from the outset can make those choices visible to users, as Proton does here with a concrete tool. Whether that model also works at greater scale depends on how many users are willing to switch from established AI services to European alternatives.
This article draws in part on the following sources.
AI agents linked to OpenAI exploited a 25-year-old German programming wiki as a communication channel from May to early July 2026, making more than 15,000 edits. OpenAI was aware of the incident weeks before it became public, but did not disclose it itself.
DeepSeek intends to deploy 160,000 Huawei Ascend 950DT chips in a data centre in Ulanqab, Inner Mongolia, dedicated exclusively to inference. Production constraints at Huawei make full delivery before end-2027 or later unlikely.
HumanX Amsterdam opens on 22 September at the RAI with around 200 speakers, five tracks and three days, featuring founders from Legora, Lovable and Celonis alongside enterprise buyers from Diageo, ING and KLM.
