Engineering · Security
Every pull request gets a security review before a person merges it.
An AI reviewer reads each change for security issues, with read-only access to the code. Findings go into your issue tracker; merging stays with your engineers.
Book a free strategy callHow does an AI security review on every pull request work?
When a pull request opens, an AI security reviewer reads the change with read-only access to the repository, alongside automated checks for secrets, vulnerable dependencies and risky code. Findings are posted on the pull request and filed in your issue tracker with severity and a suggested fix. The reviewer can't merge, push or change settings. A person decides and merges. It already runs in production.
- Who approves
- The engineer who merges, and your tech lead
- You own
- The workflow, agent logic, integrations, evals and logs
When AI writes code faster than anyone can review it
AI coding tools produce more changes than a small team can review line by line. Every line is something to test, secure and maintain, so the check has to run on every change, before it reaches production.
“Half our pull requests are written with AI now, and review is the bottleneck.”
“We found an API key in the repository three months after it was committed.”
“Nobody on the team is a security specialist, and every change touches login or payments somewhere.”
From a pull request to a merge decision
- 1
A pull request opens
The review starts on its own for every new pull request and every update.
- 2
Automated checks
Secrets, vulnerable dependencies and known risky code patterns are checked first.
- 3
AI security review
The AI reviewer reads the change in the context of the code around it, looking for issues such as missing access checks, injection and unsafe handling of data.
- 4
Findings filed
Each finding is posted on the pull request and filed in your issue tracker with severity, location and a suggested fix.
- 5
A person merges
The engineer and the reviewer on your team decide what to fix, and merge. The AI has no permission to merge.
What the AI can touch, who approves, what gets logged
Each solution runs with the narrowest access that does the job. Drafts stay drafts until a person sends them.
It can
- Read the code and the pull request
- Comment on the pull request
- Create issues in your issue tracker
It can't
- Merge, push or approve a pull request
- Change repository settings or secrets
- Deploy anything
- Read repositories outside its scope
- Who approves
- The engineer who merges, with your tech lead deciding which findings block a release.
- What's logged
- Each review, the findings and their severity, the issues created, and who merged after which findings.
In production today, and what it changed
It runs in production on real repositories, on every pull request. For a one-off check of an app built with AI tools, there is the Vibe-Code Security Audit; for a team adopting AI coding tools with test gates, the AI Engineering Method.
Measured on your data, before and after
What we measure, before and after: Security findings caught before merge against issues found after release, and review time per pull request.
What you provide, what we set up, how long it takes
You provide
- Access for a review app on the repositories in scope
- Your issue tracker
- Your rules: what blocks a merge, what is only a warning
- One tech lead who owns the gate
We set up
- The review app, with least-privilege permissions
- The automated checks and the AI review prompts, with tests (evals)
- Findings filed in your issue tracker, with severity
- The log and the before/after measure
Embed time
It starts in comment-only mode on a few repositories, then becomes a required check.
What we adapt to your company
The review loop stays the same. Your stack and your thresholds shape it.
- Your code host and issue tracker
- The languages and frameworks in your stack
- What blocks a merge and what is only a warning
- Which repositories are in scope
When you need more than the solution
Questions before you start
Bring a recent pull request.
On a strategy call we look at your repositories, your review process and where AI-written code ships without a security check. If the solution doesn't fit, we'll tell you on the call.