Cursor Security Review and Rollouts: Setup and Origin Submissions
Cursor Security Review is a pull request security reviewer that Cursor launched on September 23, 2026, together with Rollouts. It comments once on each PR and reports only exploitable vulnerabilities, with a severity level, an attack path and a suggested fix. Both features are currently available on the Teams and Enterprise plans. On October 13, the Origin app marketplace also opened to public submissions. Everything below is accurate as of October 2026 and based on the official Cursor changelog.
What does Security Review actually do?
According to Cursor, it reviews each PR using the context of the whole codebase and leaves a single comment on the PR. Coverage includes:
- Injection (SQL, command, template and similar)
- Authentication and authorization bypass
- Secrets committed to the repository
- SSRF and open redirects
- Unsafe deserialization
- Dependencies with known vulnerabilities
Each finding comes with a severity level, an attack path and a suggested fix. Cursor describes it as looking for exploitable vulnerabilities, not code style issues.
Two details to note. Draft PRs are skipped, so you will not see scan results while a PR is still a draft. Teams can also define custom codebase rules, and every PR is checked against them.
What is Rollouts?
Rollouts covers the last mile after merge. When a PR is opened, it posts a monitoring plan as a comment, listing the risks it identified, the expected impact, the signals it will check and any instrumentation gaps. You can edit the plan directly, and it will use your edited version.
After the code is deployed, it gives one of three verdicts per deployment environment: verified healthy, regression detected, or inconclusive. When it detects a regression, it points to the suspect change and notifies the author. Depending on configuration, it can open a revert PR for a human to review, or hand the issue to a cloud agent. Cursor states explicitly that Rollouts does not merge or roll back on its own.
Setup requires three kinds of systems: source control (Origin or GitHub), a deployment system that supplies deploy events, and telemetry (such as Datadog). Feature flag integration is listed as "coming soon."
How do you turn it on, and what permissions does it need?
Per Cursor's documentation, Security Review is enabled for selected repositories in the dashboard, and Rollouts is enabled in the automations tab. The exact entry points depend on Cursor's current UI. Cursor also mentions a limited-time trial allowance, roughly 50 changes for Teams and 500 changes for Enterprise, lasting 10 days. Check Cursor's official pages for billing and quotas.
You can work through this checklist in order:
| Step | What to check |
|---|---|
| 1 | Enable Security Review on one or two non-critical repositories first, rather than rolling it out org-wide |
| 2 | Confirm the repository authorization scope: grant access only to those repositories, not all of them |
| 3 | Before enabling Rollouts, check whether the deployment system and telemetry accounts have read-only permissions; create a dedicated service account if needed |
| 4 | Write one to three custom rules, for example your team's known internal auth conventions |
| 5 | When you decline a finding, enter a closing reason (per Cursor, a finding closed with a reason will not be raised again on the same PR) |
| 6 | After a week, count false positives versus valid findings before deciding whether to expand |
How should you think about false positives and false negatives?
The Cursor changelog gives no false positive rate, so no numeric judgment is made here. In practice, we suggest the following:
- Treat it as one more check, not as a security verdict. It comments once per PR and covers only the categories Cursor lists. Business logic flaws and configuration issues still need human review and your existing SAST and dependency scanning.
- When it reports something as exploitable, reproduce it by following the attack path it gives before deciding whether to fix it.
- When Rollouts returns inconclusive, look at the instrumentation gaps and the signals listed in the monitoring plan to see whether part of your service is not wired up.
What does the Origin marketplace opening to submissions mean?
On October 13, the Origin app marketplace added more than 20 developer tools. Cursor lists partners including Cloudflare, Linear, Sentry and Socket Security, as well as Vercel, Netlify, CircleCI, CodeRabbit and Semgrep, covering deployment, CI, code review, security, monitoring and project management. Public submissions are now open as well.
For teams that want to submit their own app, the official process is:
- Create the app under your own namespace and keep it private while you build.
- Build on the Origin API. Cursor describes apps as using an app JWT and installation access tokens, receiving webhook events, and writing check results, PR comments and reviews. Refer to the official Origin API documentation for details.
- Once the app is public, submit the slug, suggested tags and a demo video on the app page. The video must show install, uninstall and the core flow.
- After Cursor reviews and approves it, the app is listed.
The publishing guide and the Origin API reference are linked from that entry in the official changelog. As a user, review the permission scope a third-party app requests before you install it.
What changed at GitHub Copilot at the same time?
In its October 2 changelog, GitHub announced that Copilot code review can be triggered through the REST and GraphQL APIs, with the review effort level configurable per request. The "Default" level now maps to Balanced (effective September 28); repositories or organizations that had explicitly chosen Lite are unchanged. The feature is available on Copilot Pro, Pro+, Max, Business and Enterprise. The two vendors take different approaches, and this article does not rank them. One practical note: if your team uses both, make sure the same PR does not receive several overlapping comments.
Related reading and tools
- Other recent Cursor features: Cursor can now remote-control local agents from your phone
- Isolating agents before they run commands: Copilot local sandboxing is now generally available
- Rate limits or degraded service on Cursor: Cursor down or hitting frequent 429s? October degradations
- Data handling terms for AI tools: How to judge an AI API's data privacy policy
- To check whether the service is currently up, see /en/down/cursor. Official update records are at /en/changes.
