AI Doesn’t Have to Mean Giving Up Privacy: What Proton’s CEO Wants You to Know

Intro

Every time you paste a paragraph into ChatGPT, ask Claude for a summary, or let Copilot draft an email, you hand over data. For many people, that trade‑off feels inevitable – you can’t use modern AI tools without exposing your personal information. But Proton’s CEO, Andy Yen, argues that privacy in the AI era is actually possible. There’s one thing, though, that keeps him up at night.

What Happened

In a recent interview with Spiceworks, Yen laid out his view that people don’t have to surrender privacy to benefit from AI. Proton – the company behind encrypted email, VPN, and cloud storage – has been building its own privacy‑focused AI features. The CEO acknowledged that AI can be built in a way that respects user data, but he also pointed to a single, persistent risk that still bothers him: data centralisation.

According to the article, Yen’s primary concern isn’t the technology itself but how it’s controlled. When a handful of large companies hold most of the training data and the infrastructure, users have very little say in what happens with their information. That lack of transparency and user control is, in his view, the biggest unresolved problem.

Why It Matters

The numbers back up the worry. Adoption of generative AI tools has surged, but surveys consistently show that a majority of users are uneasy about how their data is collected and used. Much of that unease is justified. Many popular AI services store prompts, often for model improvement. Some share data with third parties. And because the models are usually run on centralised cloud servers, the provider has access to everything you type.

This creates a few concrete risks:

  • Your conversations can be reused. Even if a service promises not to read your data later, metadata or the content itself may be used to train future models – potentially exposing patterns or personal details.
  • A single breach can expose everything. If a centralised AI provider is compromised, attackers can get years of your queries, documents, and summaries.
  • You have no way to verify what happens behind the scenes. Most companies publish privacy policies, but they are often vague, and auditing is nearly impossible for an individual user.

These risks aren’t hypothetical. Already, employees have accidentally leaked trade secrets by pasting confidential data into public AI tools. On a personal level, people’s health questions, financial details, or private thoughts become part of a permanent record they cannot delete.

What Readers Can Do

The good news is that you don’t have to stop using AI altogether. Several practical steps can reduce your exposure.

1. Favour services with strong privacy defaults. Proton itself now offers an AI assistant that runs on its encrypted infrastructure. Other companies like Apple have opted to process as much AI work as possible on the device. When you must use a cloud-based tool, check whether it promises not to use your data for training and whether processing happens in‑memory rather than stored on disk.

2. Avoid sharing sensitive information unnecessarily. Even with a trusted provider, be mindful of what you type. Treat a conversation with an AI assistant the way you would a semi‑public chat: no passwords, no medical records, no intimate details you wouldn’t want repeated.

3. Use local or open‑source models when possible. Tools like Llama (via Ollama) or Mistral can run entirely on your own computer. Once downloaded, no data ever leaves your machine. The convenience may be slightly lower, but the privacy gain is enormous.

4. Limit the number of AI services you use. Every new account is another potential leak point. Stick to one or two services that have proven privacy practices, and avoid using AI features embedded in apps that don’t need them (e.g., a note‑taking app that sends your text to the cloud for AI summarisation without asking).

5. Review settings regularly. Many AI tools allow you to delete your chat history or opt out of model training. These options are often buried in account settings. Spend five minutes checking yours, and set a reminder to do it again every few months.

6. Consider a privacy‑focused VPN. While a VPN doesn’t protect the content of your prompts, it can prevent your internet provider or anyone on your local network from seeing that you are connecting to an AI service. Proton’s VPN, for what it’s worth, is one option – but any reputable no‑logs VPN will help.

Sources

The information in this article is based on the Spiceworks report “Privacy in the AI era is possible, says Proton’s CEO, but one thing keeps him up at night” (published June 4, 2026). Additional context comes from Proton’s public documentation and widely reported surveys on AI privacy concerns. The specific concerns about data centralisation attributed to Andy Yen are paraphrased from that interview and align with his past public statements on the topic.