How to Use AI Without Sacrificing Your Privacy, According to Proton’s CEO
Artificial intelligence tools are now embedded in everything from email to photo editing, and they often ask for a lot of data to work well. Many users worry this trade-off is unavoidable. Proton’s CEO, Andy Yen, disagrees – but he also points to one risk that keeps him up at night.
Here is what he said in a recent interview, and what it means for anyone trying to stay private while using AI.
What happened
In an interview published by Spiceworks on June 4, 2026, Yen said that privacy in the AI era is possible if you choose the right tools and habits. Proton, best known for its end‑to‑end encrypted email, VPN, and cloud storage, has been expanding into privacy‑focused AI features.
But Yen also admitted that one thing keeps him up at night: the risk that users feed sensitive personal or business data into AI systems that lack proper encryption or data handling policies. Even if an AI tool itself is safe, a single prompt containing a credit card number, a medical record, or a confidential work document can leak through third‑party integrations, logging practices, or model training pipelines. The threat is not the AI model itself – it is the data you give it.
Why it matters
Most consumers do not realise how much of their data is collected when they use free AI chatbots or image generators. A 2025 survey by the Pew Research Center (cited in the same article) found that 67% of Americans say they understand little or nothing about how companies handle their data when they use AI tools. Meanwhile, companies routinely use prompts to train models, store logs for months, or share data with third‑party processors.
The problem extends beyond privacy violations. If an AI model is trained on your private conversations without your consent, that information can reappear in responses to other users – a well‑documented risk in large language models. Once your data is out, you have little control over it.
Yen’s core argument: privacy in the AI era is not impossible, but it requires being deliberate about which services you trust and how you use them.
What readers can do
Proton’s CEO recommends several concrete steps, many of which align with broader privacy best practices:
1. Use AI tools with end‑to‑end encryption. If a service cannot read your data, it cannot leak or misuse it. Proton offers encrypted AI features in its products, and other providers are starting to follow suit. Look for terms like “zero‑access encryption” or “on‑device processing.”
2. Avoid sharing sensitive information in prompts. Think of every prompt as a permanent record. Do not paste passwords, financial details, or personal identification numbers into any AI chat – even if the service claims to delete them.
3. Audit app permissions. Many AI assistants run on phones or browsers and have access to your location, contacts, or camera. Review these permissions regularly and revoke anything that is not strictly necessary.
4. Use privacy‑focused browsers and search engines. Firefox, Brave, or DuckDuckGo reduce tracking. Some also block AI‑driven data collection scripts.
5. Consider using a VPN. When you use public networks or untrusted Wi‑Fi, a VPN encrypts your connection so that your AI tool usage is not visible to your internet provider or potential snoops.
6. Read privacy policies – or at least the data retention section. If a service says it retains prompts for “model improvement,” assume your data will be used indefinitely. Look for services that promise not to train on user data.
None of these steps are foolproof, but together they shrink the surface area for data leakage. As Yen put it in the interview, “privacy is not a binary state – it is a series of choices.”
Sources
The original article is available at Spiceworks: “Privacy in the AI era is possible, says Proton’s CEO, but one thing keeps him up at night” (June 4, 2026). The interview also references a 2025 Pew Research Center survey on AI and data awareness. Proton’s official blog and product documentation provide further details on their encryption practices.