Cursor ships Security Review + Rollouts agents for the last mile
Two bots scan pull requests for injection and auth-bypass bugs and watch prod deploys, reverting regressions — as GitHub widens Copilot review triggers.

Copy markdown
Security Review flags exploitable PRs, not style nits
Cursor's new Security Review agent scans each pull request for real exploits — SQL, command and template injection plus authentication and authorization bypasses — and accepts per-repo custom rules. It runs on Teams and Enterprise plans with trial credits, so a security pass can fire on every PR before a human reviewer touches it.
Rollouts watches your deploy and can roll it back
The companion Rollouts agent monitors a PR's deployment across environments and returns a verdict — verified healthy, regression detected, or inconclusive — building a monitoring plan and triggering reversions when it spots trouble. It's an agent babysitting production so a bad merge doesn't sit live while you're offline.
GitHub widens the triggers for Copilot code review
GitHub shipped more ways to request and configure Copilot code reviews on Sep 23: finer repo- and org-level rules for when Copilot auto-reviews a PR. It's the same push to automate the review-and-ship last mile, and it's free to wire into your existing PR flow if you already have Copilot.
Elsewhere: Alibaba's Agent Context claims 67% fewer tokens
At its Sep 24 full-stack AI event Alibaba launched AgentCore, an enterprise platform to build and run agents, alongside Agent Context — persistent memory plus real-time retrieval it says trims token usage up to 67% in knowledge-heavy workloads. Worth a look if you build agents on Alibaba Cloud.