On September 23, 2026, Cursor announced "Rollouts and Security Review" in its official changelog: one capability for continuous monitoring after code ships, and another for security review before a PR is merged. Launching both at once extends the programming agent's remit from "writing code" to "everything after delivery."
Rollouts is straightforward in concept: it automatically generates a monitoring plan for each PR and keeps watching key metrics once the code is deployed. When it detects a regression, it catches the anomaly before users notice, pinpoints the suspicious change, notifies the author, and can open a revert PR or hand the issue to a cloud agent. It never auto-merges or auto-rolls-back — the final decision stays with engineers.
Rollouts: filling the post-deployment monitoring gap
On the integration side, it works with Origin or GitHub for version control, pulls deployment events from continuous delivery (CD) systems, and connects to telemetry services like Datadog for metric signals; a feature flag integration is still on the way, described as "coming soon." Agents used to be great at writing code and opening PRs but absent once the code actually ran — Rollouts fills exactly that gap.
Security Reviewer: a clear division of labor with Bugbot
Security Reviewer is a standalone security review bot: on every PR, it scans the full codebase context for exploitable vulnerabilities and reports everything in a single comment. The division of labor is crisp — security issues go to Security Reviewer, code style and quality stay with Bugbot, and the two bots don't step on each other's toes.
The scan coverage is substantive: SQL, command, and template injection; authentication and authorization bypass, including checks silently broken by a refactor; secrets and credentials accidentally committed to the repo; SSRF and unvalidated redirects; insecure deserialization; and dependency changes that introduce known vulnerabilities. The review traces where user input enters, which paths it travels, and where it lands, with each finding carrying a severity, an attack path, and a suggested fix. Teams can also set team rules — for example, which client external calls must go through, or which tables may not be queried directly in request handlers. Once you dismiss a finding with a reason, it won't nag you again in the same PR.
Impact: what it means for enterprise teams in practice
Both capabilities are available to Teams and Enterprise plans, enabled in the automations tab. The company says scanning is 21% faster, with average time down from 4.8 to 3.8 minutes; third-party reports put security comment adoption at roughly 60–70%. There's a free trial quota for 10 days after launch — about 50 changes for Teams, 500 for Enterprise — a window for smaller teams to validate the effect at low cost. For engineers responsible for production stability, regressions caught before users notice means part of the on-call pressure gets absorbed up front.
Our take
Notably, news broke not long ago that OpenAI plans to end its model-supply contract with Cursor. Shipping post-deployment monitoring and security review as default capabilities at this moment answers the question enterprise customers care about most: who's responsible when something breaks. Rollouts folding the "after launch" phase into the agent workflow signals that an agent's value is shifting from "writes fast" to "doesn't break things"; Security Reviewer existing as its own bot shows security review moving from a manual checklist to a default pipeline step. The industry is confirming the direction: AI coding tools are racing from "writing code" toward "everything after delivery," Anthropic's Claude Security is likewise in enterprise public beta, and code security agents are becoming an enterprise standard.
