Machine Learning Engineer

Join Proton and build a better internet where privacy is the default

Proton was founded in 2014 by scientists from CERN on a simple truth:
privacy is a fundamental human right
. Since then, we’ve built the world’s largest encrypted email service (Proton Mail) and expanded into Proton VPN, Proton Drive, Proton Pass, and Proton Calendar—tools used by millions globally to protect their freedom, fight censorship, and keep their data safe. In some situations, Proton has literally helped save lives!

We are profitable, independent (no VC control), and selectively hire from the top ~1% of applicants. Our 500+ team members across 50+ countries come from leading organizations and elite academic backgrounds. We move fast, keep hierarchy light, and prioritize impact over optics. If you want to do meaningful work with exceptionally high-caliber people, this is it. Join us and do work you can truly be proud of.

Check our open‑source projects here!

The MSA team (Mail Delivery, Spam, and Anti-Abuse) is a multidisciplinary group founded in 2019 to solve complex security challenges across the Proton ecosystem. We build sophisticated systems from scratch that combine human intelligence and machine learning to make tens of millions of real-time or asynchronous decisions daily. Our focus areas include:

Over the past few years, our custom systems have:

Reduced spam filter misclassifications by over 70%

Blocked millions of abusive bulk signups

Protected hundreds of thousands of users from account compromises

Mitigated hundreds of DDoS attacks

Recently, we’ve expanded our impact with Proton Sentinel (an AI-human hybrid security program) and Proton CAPTCHA (our own puzzle system to detect and prevent abuse). We’re now scaling our capabilities with agentic AI to enable autonomous threat detection and response.

In 5 years, we’ve grown from 2 engineers to 40+ engineers and analysts across 3 continents, operating 24/7. We’re looking for curious, collaborative, and impact-driven engineers who thrive in a startup environment and want to shape the future of secure AI.

Design and deploy scalable ML systems using modern MLOps practices, including real-time inference, model monitoring, and automated retraining pipelines.

Architect agentic AI systems capable of autonomous threat detection, adaptive policy enforcement, and self-healing security responses.

Build tools for model development
, debugging, and explainability to accelerate iteration and transparency.

Optimize ML workflows for distributed environments to handle large-scale data processing.

Collaborate cross-functionally with security analysts, backend/frontend engineers, and customer support to align technical solutions with user needs.

Advance the state of ML for anti-abuse and account security
, exploring cutting‑edge techniques like adversarial robustness, graph-based anomaly detection, and privacy‑preserving AI.

Degree in Computer Science or a related quantitative field, with 2+ years hands‑on experience in building and running ML systems.

Production-level ML expertise: You’ve shipped models that handle real-world scale and complexity.

Hands‑on experience with LLMs in agentic environments.

Strong software engineering skills: Proficiency in Python, with a solid understanding of backend and server fundamentals.

Statistical rigor: Deep understanding of probability, hypothesis testing, and experimental design.

• Experience with adversarial ML or red‑team­ing ML systems.
• Experience working on distributed systems
  .
• Contributions to open‑source security/privacy tools
  .
• Knowledge of federated learning, differential privacy, or on‑device inference
  .
• Familiarity with graph‑based anomaly detection or knowledge graphs
  .
• Technical leadership: Experience owning projects, mentoring peers, and driving technical direction.
• Security mindset: Familiarity with abuse patterns, adversarial attacks, or threat intelligence (bonus points!).

• Impact at scale: Your work will protect millions of users globally.
• Cutting‑edge tech: Tackle challenges like agentic AI, real‑time…