Cursor Rollouts and Security Review are two new team-focused bots for the last mile of software delivery. Rollouts watches deployment health for a pull request, while Security Review looks for exploitable bugs in the pull request itself. This page treats them as workflow tools, not as a promise that an AI review catches every regression or vulnerability.
What the two bots do
Rollouts reads a change and the systems it touches, then writes a monitoring plan as a pull-request comment. After a deploy event, it evaluates logs, metrics, and traces per environment and reports verified healthy, regression detected, or inconclusive. If it finds a regression, Cursor says it can notify the author and, depending on configuration, open a revert pull request for review or hand the finding to a cloud agent. It does not merge or roll back on its own today. It connects to Origin or GitHub, a continuous-delivery system, and telemetry providers such as Datadog.
Security Review reads every pull request in the context of the codebase and posts one comment focused on exploitable bugs. Cursor lists injection across SQL, command, and template surfaces; authentication and authorization bypasses; leaked secrets; SSRF; unsafe deserialization; and risky dependency changes. Team rules can add organization-specific constraints. Draft pull requests are skipped, and the release is available on Teams and Enterprise plans.
Best fit and not fit
Best fit: engineering teams that already have source control, deployment events, telemetry, code ownership, and a human approval process for merges and reverts.
Not fit: teams without reliable instrumentation, repositories with no review owner, or organizations that need a security certification or complete vulnerability coverage from one automated comment. A bot finding is a signal to investigate, not proof that the code is safe.
Our editorial decision rule
Start in one non-critical repository. Compare Rollouts’ verdicts with existing tests and incident signals, and compare Security Review findings with your current scanner or manual review. Record false positives, missed issues, telemetry gaps, time to investigate, and whether a proposed revert is safe. Keep human approval for merges, reverts, production changes, and changes to authentication or secrets. Choose these bots when the evidence loop reduces review effort without weakening ownership; choose a conventional scanner or manual threat model when the integration cannot be audited.
Evidence and limitations
The official Cursor changelog supports the feature scope, integrations, plan availability, and stated limitations. It does not establish detection rates, uptime, or security guarantees. We have not independently tested Rollouts, Security Review, telemetry integrations, or usage credits. Read the Cursor profile, compare Cursor vs Replit, and follow the AI code review guide before enabling the bots on a production repository.nn
October 19 update: incomplete Security Reviews no longer count as clean
Cursor’s web release notes for the week of October 19 say that when part of a Security Review fails, the review no longer treats the result as clean or auto-approves the pull request. The same release notes describe improvements to how findings are retained at the deadline, transient connection errors are retried, and an alternate model is used when the selected model is unavailable. Cursor also says open findings in unchanged files now count toward the security check, and multi-file findings are re-checked after a push.
This is a material safety behavior, not just a dashboard change: a failed or partial review should be visible as incomplete rather than interpreted as evidence that the pull request has no security issues. Teams using Cursor Security Review should confirm the check’s required status in their repository rules, inspect how failed or timed-out reviews appear in CI, and keep a human security review for changes with sensitive data or high impact. A tool’s fail-closed behavior reduces one class of false reassurance; it does not establish that the scanner detects every vulnerability.
What this does—and does not—prove
Cursor’s changelog documents a product behavior change but provides no independent measurement of detection quality. Security Review is an aid for finding exploitable patterns and reviewing pull requests, not a certification, penetration test, or substitute for secure development practices. Before relying on it as a required control, test a deliberately incomplete review and verify that your branch policy blocks merging as expected. Also confirm that exception and dismissal workflows are auditable and that ownership is clear.
Teams evaluating broader delivery automation can compare these controls with Replit’s monitoring and auto-republish workflow; the products act at different points in the software lifecycle. Evidence note: This update is based on Cursor’s official release notes for the week of October 19, 2026; AI Tools Manual did not independently test the security agent. Next review: November 11, 2026.