Proton's Yen stops refusing AI and ships Lumo, moving the privacy fight to terms

Proton released the latest version of Lumo, its own AI chatbot, in late June, and founder and CEO Andy Yen told Wired he is not anti-AI any more than he is anti-internet ^[1][2]. For a company whose product line is the end-to-end encrypted counterpart to Gmail, Docs, Sheets, Calendar and Meet, that is a concession with consequences: refusal is off the table, and the argument shifts to terms and architecture ^[3].

Wired's Andy Greenberg frames Yen as among the most influential voices in the industry equating AI with privacy invasion ^[4], and lays out the standard objection alongside it: these systems sift vast troves of everyone's data to produce their results, and consumer chatbots encourage users to share their deepest secrets with a cloud-hosted large language model ^[5]. Yen's own commercial read is that the push to integrate AI into every consumer product, whether people want it or not, is driving new users to Proton in droves ^[6]. That is the uncomfortable part of the position. The default that fills his funnel is now a thing he also ships.

The stated answer, per Wired, is that Proton is integrating AI more deeply into its suite while trying to do so in a way that is more voluntary and preserves privacy far more than the typical tech giant does ^[7]. Wired describes this as a willingness to embrace an apparent contradiction between AI and privacy, not a claim to have dissolved it ^[8]. "Voluntary" is a commitment about product surfaces: whether the assistant is opt-in or stapled into every compose window. "Preserves privacy" is a claim about infrastructure, and this is where the published excerpt stops short. It does not say where Lumo's inference runs, what is retained, for how long, or under whose jurisdiction ^[9]. Naming the product is the start of the disclosure, not the whole of it.

Yen's founding argument is what raises the stakes on retention. He told Wired that the realization behind Proton was that government surveillance is a problem but corporate surveillance is the bigger one, and that US authorities do not really need the NSA when they can directly subpoena Google and Facebook, which he says they do routinely ^[10]. Applied to chat logs, that is an argument about what a vendor keeps rather than what it promises: data that exists can be compelled. Proton Mail launched in 2014 out of that thinking, conceived during a Large Hadron Collider upgrade cycle in the summer after Edward Snowden's 2013 disclosures ^[11][12].

Watch for whether Proton publishes inference location and retention terms for Lumo in the same specific register it uses for mail encryption, and whether the assistant stays opt-in per surface as the company pushes AI ever more deeply into the suite ^[7]. Watch also whether the signup pattern Yen describes survives contact with a market in which every privacy vendor ships an assistant ^[6], and what he says about the best corporate structure for a tech company not solely designed to maximize profit, a subject the interview covers ^[13].