/

/

Security-Only Reviews: A Security Reviewer on Every Pull Request

Security-Only Reviews: A Security Reviewer on Every Pull Request

AI Code Review

Amartya | CodeAnt AI Code Review Platform

Emroz Ahmed

Product Marketing

CONTENTS

0% READ
Ship Better Code, Faster

Catch bugs and accelerate code reviews with AI.

Introducing CodeAnt AI Security-Only Reviews: A Security Reviewer on Every Pull Request

Turn on security-only reviews and CodeAnt AI checks every pull request for security issues, with review suggestions limited to security findings.

Today we’re launching security-only reviews in CodeAnt AI. Turn them on for your organization or for a single repository, and CodeAnt AI reviews every pull request for security issues. Its review suggestions cover security findings and nothing else.

Security-only reviews cover 15 categories of security finding, from SQL injection and command injection to authentication bypass, authorization bypass and hardcoded secrets. Each finding lands on the line that introduced it, with what’s wrong, the steps that would trigger it and a fix your developer can commit. Style notes, naming and non-security bugs stay out of the suggestions. Other parts of the review, such as pull request descriptions and quality gates, keep following their own settings.

If your organization uses Code Security without AI Code Review, security-only reviews are already on. Authors with a Code Security seat get them on their pull requests automatically.

For an engineering leader, the result is simple: every pull request gets a security review, including the ones your security team never sees, and developers hear about a security problem while the code is still open in their editor.

Security Only Reviews switched on for payments-service, a repository of Northwind Labs, a fictional company.

Where it works best

Teams with a review process they already trust. If your senior engineers or another tool already cover style and design, a second set of opinions on naming adds noise. Security-only reviews add a dedicated security check to every pull request without adding opinions on everything else.

Security teams outnumbered by developers. An application security team can’t read every pull request. Security-only reviews put a security reviewer on each one, so your security engineers can spend their time on design reviews, threat models and the findings that need a person.

Codebases with a lot of AI-written code. Coding assistants raise the volume of code faster than anyone can review it for security. A reviewer that reads every change for injection, broken access control and leaked secrets keeps pace with the volume.

Code Security customers. Organizations on a Code Security plan without AI Code Review get security-only reviews for authors with a Code Security seat. Those reviews post security suggestions only, with no pull request description or merge-risk comment, and quality gates run as usual.

A first step into AI review. Some teams want to see an AI reviewer earn trust before it comments on everything. Starting with security-only reviews gives developers a small number of high-stakes comments, and you can widen the review repository by repository.

Where full review fits better. If you want logic errors, performance problems and maintainability flagged too, keep full AI review on and use the severity threshold to control how much reaches each pull request.

How it works

The screenshots below are from CodeAnt AI, configured for payments-service at Northwind Labs. The pull request comment is an illustration.

1. Switch it on. In Settings, open AI Code Review Configuration, pick the organization or a repository, and set Security Only Reviews to On under Review Behavior, as shown above. It is off by default, and a repository can override the organization’s setting either way. It applies to new reviews from then on.

2. CodeAnt AI keeps only security findings. Each review still reads the whole change, then keeps the findings in its security categories: SQL injection, command injection, SSRF, cross-site scripting, CSRF, path traversal, insecure deserialization, XXE, authentication bypass, authorization bypass, hardcoded secrets, sensitive data exposure, insecure cryptography, open redirects and other security issues. Everything else is filtered out, and Nitpicks are limited to security issues too.

3. Developers see the problem, the proof and the fix. Each finding is an inline comment on the changed line. It says what’s wrong, offers a fix the developer can commit in one click, lists numbered steps of reproduction that show how the issue would be triggered, and includes a Prompt for AI Agent with the full context for Cursor, Copilot or Claude Code. Critical and high severity findings also explain their downstream impact.

Illustrative: a security-only review flags a missing ownership check on a new payout endpoint in Northwind’s payments-service.

4. It works with your noise controls. The security filter runs first, then your Suggestion Threshold and Nitpicks settings. With the threshold at Major, only Major and Critical security findings appear as suggestions. With Nitpicks on, minor security notes are gathered into one collapsed comment.

payments-service runs security-only reviews with the Suggestion Threshold set to Major.

Why it matters

Most security bugs arrive in an ordinary pull request. In August 2026, our security research team disclosed nine vulnerabilities in Dolibarr, an open-source ERP and CRM that thousands of small and mid-size businesses use to pay their suppliers. In one, an ordinary user with the routine right to create company records could change a supplier’s bank details, even though the app wouldn’t let them view that supplier. The next payment file looked normal, except for the account the money went to (CVE-2026-71507, fixed in Dolibarr 24.0.0). The Northwind example above is the same class of bug: one handler, one missing check, written in a single change.

Human review catches some of these and misses others, and the misses follow a pattern. A study of Chromium OS compared 516 code reviews that caught a security defect with 374 that missed one. Reviews were less likely to catch the defect when the change touched more directories and when the reviewer had done more reviews overall (ICSE 2021). Big changes and busy reviewers are exactly what fast-moving teams have.

AI-written code adds volume. In Veracode’s 2025 tests of more than 100 language models on 80 coding tasks, the generated code introduced a security vulnerability in 45% of cases (Veracode). And the bill for what slips through is real: IBM puts the average data breach at $4.44 million worldwide and $10.22 million in the US (CyberScoop).

What changes for your organization

Every pull request gets a security review. Including the late-night fix, the small refactor and the change your security team never hears about.

Developers fix security issues while the code is fresh. The finding arrives on the line they just wrote, with the steps to reproduce it and a fix to commit.

Your security team focuses on the hard problems. Routine checks happen on every pull request, so people spend their time on design, threat modelling and the findings that need judgement.

You control the rollout. Turn it on for one repository, a group of them or the whole organization, and change your mind at any time.

Get started

Security-only reviews are live now in CodeAnt AI. Open Settings, then AI Code Review Configuration, choose the organization or a repository, and set Security Only Reviews to On. Read the code review documentation for the full list of categories and how the setting works with Code Security plans.

Northwind Labs, payments-service and the pull request shown are fictional. The settings screenshots are from CodeAnt AI with Northwind example names. CVE-2026-71507 was found and disclosed by CodeAnt AI Security Research.

Frequently Asked
Questions
What are security-only reviews in CodeAnt AI?

A review mode that limits CodeAnt AI’s pull request suggestions to security issues. Turn it on for the organization or for a single repository, and every review keeps its security findings as suggestions and drops the rest, including style notes, naming and non-security bugs. Pull request descriptions and quality gates keep following their own settings. It is off by default.

Which security issues does it flag?
Does it work with the severity threshold and nitpicks?
Do we need AI Code Review to use it?
Can we turn it on for some repositories only?
Review Code Faster.
Ship With Confidence.