---
type: "post"
id: "proton-publishes-details-of-its-ai-code-review-b4156d"
slug: "proton-publishes-details-of-its-ai-code-review-b4156d"
title: "Proton publishes details of its AI code-review and security controls"
description: "Proton says AI now assists with code review, vulnerability discovery, and security monitoring, while its existing controls still apply: people approve every code change, automated security scans remain required, and engineers retain accountability."
retrieval_nugget: "Proton says AI now assists with code review, vulnerability discovery, and security monitoring, while its existing controls still apply: people approve every code change, automated security scans remain required, and engineers retain accountability. The account matters to AI builders because it describes a layered approach to using agents in sensitive development work without granting them authority to deploy."
published_at: "2026-11-16"
updated_at: "2026-11-16"
record_date: "2026-11-16"
date_kind: "published_at"
topics: ["anthropic","code-review"]
entities: ["Proton","Proton Sentinel","OpenAI","Anthropic","Schellman"]
related_patterns: ["agent-security-moves-to-runtime-boundaries"]
novelty_label: "new_fact"
what_changed: "Proton details its AI code-review practices and security controls, adding concrete safeguards to its broader AI engineering approach."
prior_refs: ["https://newruntime.com/posts/proton-cto-bart-butler-publishes-ai-proton-engineering-762636/","https://newruntime.com/posts/speed-at-what-cost-proton-the-swiss-company-b97b51/"]
source_url: "https://proton.me/blog/how-proton-keeps-your-data-secure"
source_urls: ["https://proton.me/blog/how-proton-keeps-your-data-secure"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:proton-publishes-details-of-its-ai-code-review-b4156d"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/proton-publishes-details-of-its-ai-code-review-b4156d/","markdown":"https://newruntime.com/posts/proton-publishes-details-of-its-ai-code-review-b4156d.md","json":"https://newruntime.com/posts/proton-publishes-details-of-its-ai-code-review-b4156d.json"}
markdown_url: "https://newruntime.com/posts/proton-publishes-details-of-its-ai-code-review-b4156d.md"
---

# Proton publishes details of its AI code-review and security controls

## Retrieval answer

Proton says AI now assists with code review, vulnerability discovery, and security monitoring, while its existing controls still apply: people approve every code change, automated security scans remain required, and engineers retain accountability. The account matters to AI builders because it describes a layered approach to using agents in sensitive development work without granting them authority to deploy.

## Summary
Proton says AI now assists with code review, vulnerability discovery, and security monitoring, while its existing controls still apply: people approve every code change, automated security scans remain required, and engineers retain accountability. The account matters to AI builders because it describes a layered approach to using agents in sensitive development work without granting them authority to deploy.

## What happened
In a November 10, 2026 company blog post, Proton explains how it uses AI in software development and security. It says its end-to-end and zero-access encryption foundations have not changed, and that its documented security principles—risk-based decisions, centralized governance, and explicit, controlled exceptions—still apply. Proton says it is ISO 27001 certified and completed its first SOC 2 Type II audit in July 2025, conducted by Schellman. The company says those controls cover access management, incident response, system monitoring, and risk assessment, and that another SOC 2 audit is scheduled for 2027.

For code written by people, AI, or both, Proton says the same controls apply. Every code change must be reviewed and approved by one or more people; AI reviewers are added to that process but cannot deploy code to production on their own. The reviewers required depend on the codebase: changes to core cryptography code always require review by Proton’s expert cryptography team. Automated security scans must also run on changes. Engineers remain responsible for AI-assisted work and are expected to use professional judgment and follow company policies. For major features, engineers present designs to peers before coding; the review includes security, dependencies, testing, and rollout and rollback plans.

Proton says it applies least-privilege rules to AI agents as it does to employees. Its internal Security Information and Event Management system records actions inside the security perimeter whether they come from a person or an agent, and human security analysts act on suspicious alerts. Only authorized devices can reach internal systems; authorization requires physical tokens that AI systems cannot manipulate on their own. Proton also says it monitors servers and employee laptops for indicators of compromise.

The company describes three uses of AI in security: code-change review alongside humans, continuous reviews of its own code to find vulnerabilities, and security monitoring that surfaces signals for analysts to investigate. Proton says it uses its own models and non-public cyber models from OpenAI and Anthropic for threat analysis and code scanning. Those models receive source code and threat intelligence, not customer content. For monitoring, Proton says internal AI tooling helps analysts watch a larger attack surface, while Proton Sentinel protects customer accounts. The company argues that AI is necessary to help defenders keep pace with increasingly capable attackers, while acknowledging that no company can remove every risk.

## What the archive adds
New Runtime’s earlier coverage described employee concerns about Proton’s push for faster AI adoption ([November 3, 2026 note](https://newruntime.com/posts/speed-at-what-cost-proton-the-swiss-company-b97b51/)) and an engineering memo arguing that AI changes engineers’ work toward directing models and reshaping delivery workflows ([November 3, 2026 note](https://newruntime.com/posts/proton-cto-bart-butler-publishes-ai-proton-engineering-762636/)). This is a new fact on that known story: Proton now publicly details the controls it says govern AI-assisted development and security.

## Trend signal
This supports the view that agent security depends on runtime boundaries and enforceable permissions, not just instructions to a model. Proton describes least-privilege access, audit logging, human review, and physical-token controls around internal systems. It is a company account rather than an independent assessment, but it aligns with [agent security moving to runtime boundaries](https://newruntime.com/patterns/agent-security-moves-to-runtime-boundaries/).

## Why now
Proton frames the change as a response to AI lowering the effort needed to carry out sophisticated attacks. It says defenders must use AI to keep pace, while maintaining existing controls rather than relaxing them. The post also addresses concerns about the company’s increased use of AI in engineering.

## Why it matters
The operational lesson is that deploying coding agents in sensitive environments is not simply a choice between trusting a model and banning it. Teams can let AI review changes and search for vulnerabilities while keeping deployment authority, approval, and accountability with people. Proton’s account names practical control points: approvals keyed to the codebase, mandatory scans, least-privilege access, logged actions, and human investigation of suspicious alerts.

These controls do not establish that Proton’s process catches every defect, but they offer a useful architecture for teams deciding where an agent can act and where a person must intervene. The distinction between model capability and production authority is especially important: an agent may produce findings or code without being allowed to ship either. Teams should also define what data external models receive; Proton says its external cyber models see code and threat intelligence, not customer content.

## Limits and uncertainty
This is Proton’s own description of its practices, not an independent evaluation of their effectiveness. The post gives no figures for vulnerabilities found, false-positive rates, review time saved, or incidents prevented. It does not detail the models, configurations, evaluation methods, or deployment scope behind its continuous reviews, nor explain how it validates that customer content is excluded from external model inputs. The stated controls do not eliminate risk, and Proton acknowledges that no company can do so.

## Skill to build now
Map one AI-assisted code workflow from prompt to production. Mark which repositories or files need specialist review, where automated security scans must run, who approves a change, and whether an agent can deploy. Then check agent permissions against least privilege and ensure its actions are logged. For any external model, document the exact data categories it may receive and exclude customer content unless a reviewed policy explicitly permits it.

## Reader takeaway
Proton’s stated model is to use AI broadly for review and threat detection while withholding independent production authority from agents. The reusable principle is to pair each agent capability with explicit permissions, human accountability, and observable checks—and to measure whether those controls work rather than treating a company’s policy description as proof of effectiveness.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Anthropic](https://newruntime.com/topics/anthropic.md): Explore the anthropic topic hub.
- topic hub: [Code Review](https://newruntime.com/topics/code-review.md): Explore the code-review topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)
