What Proton’s CEO Fears Most About AI Privacy — and How to Protect Yourself
Six years after the launch of ChatGPT reshaped how people interact with technology, Proton CEO Andy Yen says privacy in the AI era is not only possible but necessary. But there is one thing that keeps him up at night. In a recent interview with Spiceworks, Yen pointed to a specific risk that many users overlook while they tap into convenient AI tools.
What Happened
In an article published June 4, 2026, Yen discussed the privacy landscape around generative AI. Proton, the company behind ProtonMail and ProtonVPN, has long advocated for end-to-end encryption. Yen’s warning came at a time when AI assistants are embedded in browsers, phones, and office suites.
The particular threat that worries him is the lack of data sovereignty and persistent storage of user prompts and AI outputs by cloud-based AI services. When you type a question into a chatbot or ask an AI to summarize a document, the text of your request — along with the AI’s response — is typically stored on the provider’s servers. That data may be used to train models, shared with third parties, or accessed by employees. For Yen, this turns every AI interaction into a potential privacy leak.
Why It Matters
Most AI chatbots claim they need to store data to improve their models. But that arrangement means your emails, health questions, financial details, or proprietary business information end up on someone else’s infrastructure. Even if the AI service is “free,” your data is the price you pay — and the data may remain on servers indefinitely.
Yen’s concern is not abstract. We’ve seen data breaches at large tech companies expose user chats. We’ve seen internal leaks. And we’ve seen companies change their privacy policies after you’ve already shared sensitive information. For consumers, the risk is that a private question about a legal problem, a medical condition, or a business idea becomes part of a permanent, monetizable dataset.
The good news, Yen says, is that privacy in the AI era is achievable. But it requires a shift in how we choose and use these tools.
What Readers Can Do
1. Use AI tools that process data locally when possible. Several companies now offer offline AI assistants — from local versions of large language models like Llama or Mistral to app-based tools that run entirely on your phone without sending data to the cloud. For tasks that don’t need internet access, this eliminates the storage problem.
2. Read the privacy policy before you type. Look for services that explicitly say they do not store prompts or use them for training. Services like Proton’s own AI assistant, when available, are designed with end-to-end encryption in mind. If a policy says “we may store your data to improve our model,” assume your inputs are not private.
3. Avoid pasting sensitive information into public or free chatbots. Treat every input as if it could become public. Do not paste ID numbers, passwords, bank details, or confidential work documents into an AI chat box unless you are using a verified, encrypted service.
4. Use a privacy-focused browser extension or VPN when accessing AI tools. Even if the AI service itself stores data, reducing your digital footprint — such as masking your IP address or blocking tracking scripts — can limit how much additional information is collected alongside your prompts.
5. Ask for an option to delete your data. Many AI services now offer a way to request deletion of your chat history. If you have used an AI tool for sensitive matters, locate the account settings and ask for your data to be removed. This is not a guarantee, but it’s a step.
Sources
The information about Andy Yen’s concerns and the Spiceworks interview comes from the article published June 4, 2026, on Spiceworks: “Privacy in the AI era is possible, says Proton’s CEO, but one thing keeps him up at night.” Additional context on Proton’s encryption philosophy is drawn from Proton’s public documentation and press materials.
Note: This draft summarizes the reported statements of Proton’s CEO. For the exact wording, readers should refer to the original Spiceworks article.