The call
Approve CodeRabbit for private repositories once you set knowledge_base.opt_out where retention is a problem, install the app on the pilot repositories only, and remove linters you do not use.
Decide open-source repositories separately, because CodeRabbit's privacy policy reserves the right to train on them.
CodeRabbit sends your code to OpenAI and Anthropic for each review. Unless you opt out, it keeps encrypted caches, a code index, and the rules it learns from review chat between reviews. Its GitHub app asks for read and write access to every repository you select.
When you need this
- Security or legal must approve a third-party Git app that reads private code.
- You handle regulated data or need data kept in the EU.
- You want to know what happens to learned rules and the code index if you leave.
Worked example: the 2025 linter flaw and what it changed
What happened
In January 2025 Kudelski Security ran code on CodeRabbit's servers through a malicious Rubocop configuration in a pull request. The credentials it exposed had access to about a million repositories.
What CodeRabbit changed
CodeRabbit's incident post says it fixed the flaw within 12 hours by moving the affected linter, RuboCop, into its sandbox.
What a team controls
Install the app on pilot repositories only, and list under reviews.tools only the tools you use.
Outside contributions
CodeRabbit reads .coderabbit.yaml from the branch under review. NOFire AI added a maintainer permission check before review calls for that reason. Make the same decision for forks.
Evaluation criteria
Model vendors and training
Where does code go during a review, and is it used for training?
Look forIts FAQ says CodeRabbit never uses customer code for training. The privacy policy, updated October 16, 2026, excludes open-source projects from that commitment and says CodeRabbit uses them to train its systems.
What it keeps
What does CodeRabbit store between reviews?
Look forCodeRabbit stores encrypted code and dependency caches, a search index of your code, and learned rules. Setting knowledge_base.opt_out turns off the features that need retention and deletes stored knowledge base data. Model output logging for support is off by default, and only someone with write access to organization settings can turn it on.
Repository permissions
What can the app do in your repositories?
Look forKudelski Security's write-up records that the GitHub app asks for read and write access to code in the repositories you select. Keep it off every repository outside the pilot.
Tool execution
Which linters run on your code, and where?
Look forReview the third-party linters and security tools under reviews.tools, and turn off the ones you do not need. The January 2025 flaw ran through one of them, Rubocop. After Kudelski Security's report, CodeRabbit rotated its secrets within hours and moved the tool into its sandbox.
Configuration from the pull request
Can a pull request change how it is reviewed?
Look forDecide how CodeRabbit reviews forks and outside contributions. It reads .coderabbit.yaml from the feature branch, so a pull request author can change the review policy by editing that file. NOFire AI runs a maintainer permission check before each review call.
Region and deployment
Where does CodeRabbit run?
Look forCodeRabbit Cloud runs in a US region at app.coderabbit.ai and an EU region at app.eu.coderabbit.ai. Self-hosting keeps source code and review traffic inside your network and needs an Enterprise contract with 500 or more seats.
Access control and audit
Who can change the settings, and is there a record?
Look forEnterprise includes SSO and custom roles, plus audit logs. CodeRabbit reports an annual SOC 2 Type II audit.
Questions to ask vendors
- Which repositories need CodeRabbit, and can we install the app on those only?
- Do we set knowledge_base.opt_out, and which features do we lose if we do?
- Which configuration governs a pull request from a fork?
- Which linters and security tools run on our code, and in what sandbox?
- Are any of our public repositories covered by the open-source training clause in the privacy policy?
Red flags
- Installing the app on every repository in the organization by default.
- No record of whether knowledge base retention is on.
- Reading the FAQ's training statement without the privacy policy's open-source clause.
- Letting a review bot's approval count toward required approvals on repositories with regulated data.
Products in this piece
Sources
- Official documentation
- 11
- Public source
- 2
- CodeRabbit docs: FAQ (data security)
- CodeRabbit Privacy Policy (updated October 16, 2026)
- CodeRabbit Security
- CodeRabbit docs: Knowledge Base
- CodeRabbit docs: Architecture
- CodeRabbit docs: Configuration reference
- CodeRabbit docs: YAML configuration
- CodeRabbit docs: Self-hosted CodeRabbit
- CodeRabbit docs: Configure the VS Code extension (account region)
- CodeRabbit docs: Plans and pricing
- Kudelski Security: How We Exploited CodeRabbit: From a Simple PR to RCE and Write Access on 1M Repositories (Aug 19, 2025)
- CodeRabbit blog: Our response to the January 2025 Kudelski Security vulnerability disclosure
- NOFireAI ravel issue #1586: Retire the third-party review vendor (Sep 10, 2026)