How to Keep Your Privacy in the AI Era, According to Proton’s CEO

The CEO of Proton – the company behind encrypted email and VPN services – recently said that privacy in the age of AI is still possible, but he keeps one thing up at night. That one thing isn’t a rogue algorithm or a leaky server. It’s much simpler: the gradual erosion of consent.

In a June 2026 interview with Spiceworks, Proton’s CEO Andy Yen laid out his biggest concern: most people don’t realize how their data is used when they type a prompt into ChatGPT, Microsoft Copilot, or Google Gemini. What you share with these tools often ends up being used to train future models, unless you explicitly opt out. And the default setting is almost always “yes.”

Here’s what that means for you and what you can do about it.

What happened

Proton has long positioned itself as a privacy-first alternative to mainstream services. The company now offers a suite of tools including encrypted email, VPN, password manager, and more recently, AI-powered writing and summarization features. In the interview, Yen explained that Proton’s AI runs on its own infrastructure and is designed so that user data never leaves the user’s control – it’s processed on-device or in encrypted environments.

But the mainstream AI tools that most people use every day do not work the same way. When you paste a confidential email draft or a private journal entry into a chatbot, that text is stored on the provider’s servers. In many cases, it’s then used to fine-tune the underlying model. The catch is that this is buried in the terms of service, and ordinary users rarely read them.

Why it matters

The risk is not theoretical. In 2023, a Samsung employee was found to have accidentally leaked internal source code by pasting it into ChatGPT. More recently, journalists and doctors have reported finding their own sensitive queries showing up in public model outputs. Once your data is used for training, there is no way to unpublish it. And because large language models don’t have perfect recall, you can’t ask them to simply “forget” what you typed – the data may still be embedded in the model’s weights.

The “one thing” that keeps Proton’s CEO up at night is this: users don’t have meaningful control. Even when opt-out options exist, they are often buried in settings menus or require constant re-confirmation. The burden is on the individual, not the company.

What readers can do

You don’t have to stop using AI tools. But you can take a few steps to lower your risk.

  1. Check your settings – Every major chatbot provider has a privacy or data control page. In ChatGPT, you can disable “Improve the model for everyone” under Settings > Data Controls. For Copilot, you can turn off “Allow Microsoft to use your data to improve the model” in the privacy dashboard. Do this right away.

  2. Avoid pasting sensitive information – Treat any AI chatbot like a stranger on the internet. Do not share passwords, medical records, financial details, or confidential work documents. If you must, use a local model or a privacy-first alternative.

  3. Use encrypted or on-device alternatives – If you need AI for tasks like email drafting or summarization, consider tools that keep data on your device or in zero-knowledge environments. Proton offers a writing assistant that encrypts text before it leaves your device. Apple Intelligence processes most requests on-device. Both reduce the chance of your data being harvested.

  4. Read the privacy policy – once. It’s tedious, but spending ten minutes to understand exactly what a service does with your data can save you a lot of trouble later. Look for phrases like “may use content to train”, “share aggregated data”, or “process data in the US”.

  5. Consider running a local model – If you’re technically inclined, tools like Ollama, GPT4All, or Llama.cpp let you run an AI model on your own computer. No data ever leaves your machine. It’s not as powerful as GPT-4, but for many everyday tasks it’s sufficient.

The future outlook

Proton and other privacy-focused companies are betting that users will pay for AI services that respect their data. Whether that bet pays off depends on whether enough people realize how little control they have today. In the interview, Yen said he believes “privacy in the AI era is possible – but only if users demand it.”

For now, the safest approach is to treat every AI interaction as a public conversation until proven otherwise. The tools can be incredibly useful. Just make sure they don’t come at the cost of your personal data.

Sources

  • Spiceworks interview with Proton CEO Andy Yen, published June 4, 2026.
  • Proton’s official blog posts on AI privacy and encrypted processing.
  • European Data Protection Board guidelines on AI and consent (2024).