Privacy-first AI: The rise of confidential chatbots

The new wave of AI: privacy-first chatbots enter the scene

Proton built its name on encrypted email and VPN services, and it is now taking the same approach to AI. Lumo, its new assistant, promises to keep your conversations confidential. The launch is the latest signal of growing demand for AI tools that protect user privacy rather than just offering smart responses.

Lumo: confidential conversations in the age of AI

Lumo is different from mainstream chatbots like ChatGPT because the design puts privacy first. It uses zero-access encryption, so your conversations are stored locally on your device and stay out of reach of anyone else, including Proton itself. The company says it keeps no logs of your chats and will never use your data to train its models.

The web search feature has to be switched on manually, and even then Lumo uses only privacy-friendly search engines. It also has a “Ghost mode” that deletes your conversation after you close it, a file upload feature that analyzes documents without storing them, and Proton Drive integration for secure file sharing.

Lumo runs in the browser and as apps for Android and iOS. A free version is open to everyone, even without a Proton account, with some usage limits. For heavier use, Lumo Plus costs a monthly fee.

The broader landscape of privacy-focused AI

Lumo is not the only private AI option. Two others approach the problem differently:

• DuckDuckGo AI comes from the company behind the privacy-centric search engine. Duck.ai lets users talk to AI models from providers like OpenAI and Anthropic while keeping the conversation anonymous and untracked. It is free and requires no account.

• Kagi is a paid search engine that stays ad-free and private. Its “Assistant” feature gives access to a range of large language models for both chatting and web searching. Kagi is user-funded, so it can focus on user experience and privacy instead of advertising and data collection.

The spectrum of AI privacy

The approaches to AI privacy vary widely. On one end of the spectrum are the mainstream AI models like ChatGPT, which often use your data to train their systems. While paid versions of these services may offer better privacy policies, the free versions generally provide the least privacy.

In the middle ground sit services like DuckDuckGo AI and Kagi, which act as proxies: they anonymize your requests before passing them to other AI models. That gives you a good balance of privacy and functionality.

At the most private end of the spectrum are tools like Lumo and local, open-source AI models. These options give users the most control over their data by keeping it encrypted and stored locally.

Lumo’s launch, along with the growing popularity of other privacy-focused AI tools, points in the same direction: more users want AI that respects their privacy. As AI moves deeper into daily life, the choice of tool will likely come down to data safety at least as much as model capability.