Technology1 min read

It wasn't AI that flipped Proton's charge

By · Published by Everything Blog

In short

Proton leadership faced severe criticism after reports showed CEO Andy Yen rejected an engineer's request for days of penetration testing on new code. Yen actively pushed staff to adopt artificial intelligence and initially threatened that failure to do so could jeopardize their jobs. While the formal policy linking employment to automated tool usage was rescinded after leaking, Yen maintained it had achieved its intended purpose. A staff member later defended the directive on Reddit by clarifying that testing was acceptable as long as deployment pipelines remained on schedule.

Key points

  • Prioritizing rapid deployment over cautious, handcrafted coding introduces severe risks f…: Prioritizing rapid deployment over cautious, handcrafted coding introduces severe risks for a company centered entirely on user privacy.
  • Internal leadership maintained that penetration testing must not cause delays to the firm…: Internal leadership maintained that penetration testing must not cause delays to the firm's ongoing deployment pipeline.
  • Software flaws caused by bypassed security guardrails could result in compromised user co…: Software flaws caused by bypassed security guardrails could result in compromised user communications that prove fatal to the business.

Sources checked by our analysis pipeline: 1

I’ve been paying for Proton for many years, and it’s what I recommend when someone asks how to get off Gmail. This week 404 Media published an investigation into how Proton’s leadership is pushing AI inside the company. The detail that made the rounds came from the security Slack channel: an engineer asked for a few days to pentest new code before shipping it, and CEO Andy Yen answered that it was “bullshit and loser thinking”, then told the team to stop lovingly handcrafting code and use AI more.

For the first time, I’m not sure I’d still recommend them.

A lot of the reactions treat this as an AI story, as if any AI-written code poisons the well, and I think that misses the point. Engineers using AI to write code is the new norm. It’s fine as long as there are guardrails around it: someone reviews what the model produced, it gets tested, and someone is accountable for what ships.

What the reporting describes is a CEO telling an engineer, in front of colleagues, that raising a security…

Technology