Proton’s guide to AI privacy and security: What you should know

From surveillance to data harvesting, using AI raises some of the most important privacy questions of our time. Here’s what’s really happening to your data — and what you can do about it.

Why AI and privacy belong in the same conversation

Most AI tools are built on a trade-off: you get the help you need, they get your information. Your chats, files, and photos are used to train models, build profiles, and target ads, and in some cases, even shared with governments — often without your knowledge, let alone your direct consent.

Default settings are designed to maximize data collection, and the speed of AI adoption has outpaced most people’s understanding of what’s actually happening under the hood. This page brings together guides, explainers, and deep dives on AI privacy and security, so you can stay fully informed and in control of your data.

The risks of AI

AI privacy, data, and surveillance

When you use an AI tool, your words and files directly fund the business through model training, ad targeting, or outright sale to third parties. But surveillance doesn’t stop here. AI is also being used by governments and corporations to track, profile, and monetize people at scale — and much of it is hiding in plain sight.

Ethics of AI

AI raises hard questions about transparency in how models are built and what they’re trained on, the replacement of human labor, engagement-driven design that prioritizes profit over people, and the concentration of power in a handful of corporations. These issues shape the technology everyone is living with.

Identity and creator concerns

AI is used to replicate faces, voices, and identities for scams, phishing, catfishing, and political manipulation. At the same time, facial recognition systems and image and video generators are trained on photos and artwork scraped online without consent. The implications for personal privacy are serious and still unfolding.

AI in everyday devices

Most conversations about AI privacy focus on chatbots. But even if you’ve never signed up for an AI platform, you’re likely already feeding data into one. Some of the most data-hungry AI isn’t the kind you go looking for — it’s now built into the devices and tools you and your loved ones use every day.

AI privacy and security concerns for businesses

Businesses are integrating AI faster than they’re assessing the risks. Adding an AI assistant to your workflow often means adding a new data pipeline — one that may put your proprietary assets at risk and clash with security policies, contracts, and data protection laws. Here’s what to know before adopting AI tools for your organization.

How to protect you business files from AI security threats

Major AI platforms

The major AI assistants aren’t just tools. They’re data businesses. Here’s what you need to know about the most widely used AI platforms on the market.

OpenAI ChatGPT

The world’s most widely used AI assistant, OpenAI has faced regulatory scrutiny across Europe and North America over how it handles personal data, retains conversation history, and responds to data deletion requests.

Google Gemini

Deeply embedded in Search, Gmail, Docs, Android, and Chrome, Gemini gives Google AI-assisted access to more of your digital life than almost any other platform — raising significant questions about how that data integrates with Google’s core advertising business.

Microsoft Copilot

Copilot ships built into Windows, Office, and Teams, making it one of the most widely deployed AI tools in workplace environments. Its tight integration with Microsoft 365 has put data retention, enterprise compliance, and the controversial Windows Recall feature under the microscope.

Meta AI

Meta AI is woven into WhatsApp, Messenger, Instagram, and Facebook, where it operates alongside one of the world’s largest advertising data networks. Meta has confirmed that conversations with its AI assistant can be used to inform ad targeting.

Anthropic Claude

Claude has gained traction as the preferred AI among developers and enterprise teams, partly due to Anthropic’s stated commitment to AI safety. However, the company still does use data from both its free and paid users for AI model training by default.

Apple Siri & Apple Intelligence

Apple has built its AI features around on-device processing and its broader privacy-first brand positioning. The key question is how those commitments hold up as Apple Intelligence expands — and how much data still moves off-device for more complex tasks.

DeepSeek

The Chinese AI lab behind DeepSeek drew global attention after its models rivalled leading US competitors at a fraction of the cost. That rapid rise was quickly followed by bans in several countries and serious concerns about data routing to Chinese servers and potential state access obligations under Chinese law.

What is DeepSeek?

Grok

Grok is xAI’s AI assistant, built by Elon Musk and integrated directly into X (formerly Twitter). X’s privacy policy change in 2024 — which broadened data-sharing with third parties — and Grok’s default use of public posts for model training prompted regulatory investigations in Ireland and the UK.

Perplexity

Perplexity presents itself as a privacy-conscious alternative to Google Search, but its free and standard Pro tiers default to using your data for model training, and it tracks user activity via cookies and device fingerprinting even without an account.

Mistral

France-based Mistral markets itself as the privacy-respecting European alternative to US AI giants. However, free-tier and standard API users are opted into training data use by default — full data protection requires a paid enterprise plan.

Frequently asked questions about AI and privacy

Does AI save my conversations?: Most mainstream AI assistants do by default. ChatGPT, Gemini, and Copilot all store your conversation history and use it to improve their models unless you actively opt out. The opt-out process varies by platform and isn’t always obvious. Some platforms also retain data for a set period even after you delete your history.
Can AI companies share my data with governments?: Yes, and it has happened. Like most tech companies, AI providers are legally required to comply with valid government requests for user data in their operating jurisdictions. OpenAI, Google, and Microsoft are all US-based, meaning they are subject to US surveillance laws including FISA. Some AI providers operating in other jurisdictions — notably China — have raised additional concerns about mandatory data-sharing requirements with state authorities. In other words, as long as your conversations are stored, they can potentially be accessed.
What does AI do with my data after I’ve typed it?: It depends on the platform, but typically your input is used in one or more of the following ways: to generate your response in the current session; to train or fine-tune the underlying model; to build a profile used for ad targeting; and in some cases, to be reviewed by human contractors for quality and safety purposes. Most platforms allow some level of opt-out for training data, but opt-outs are rarely the default setting.
Is it safe to use AI for work?: It can be, but it requires care. Most consumer AI tools are not designed for enterprise-grade data security, and using them to process confidential business information — contracts, financial data, client details — may violate your company’s data policies, client confidentiality agreements, or data protection law.
Can I stop AI from using my data to train its models?: Most major platforms offer an opt-out, but few make it easy to find. Even with opt-outs enabled, some platforms retain the right to use data for “safety” or “abuse prevention” purposes.
Is there a private AI assistant that doesn’t collect my data?: Yes. Lumo is built on zero-access encryption and user-to-Lumo encryption. This means conversations are encrypted between you and the model, erased on the model’s end, and then saved in such a way that only you can decrypt your chat histories. Conversations are never used for model training, ad targeting, or shared with third parties. It’s built on the same privacy principles that underpin Proton Mail and Proton VPN.

Choose an AI that doesn’t compromise on privacy

Most AI assistants work by collecting all your inputs, storing it, and using it to improve their models or to build a profile of you. Lumo is different.

Every conversation in Lumo is confidential, so no one, not even Proton, can access your files and chats. Data is never shared with third parties, never used to train AI models, and never used to target you with advertising.

You get all the benefits of an AI assistant with none of the privacy risks, and your data stays fully yours. The way it should be.

Try Lumo now Get Lumo for Business