Vercel Security Review enters public beta, auto-scanning every GitHub push for vulnerabilities

Vercel has launched Security Review, an AI-driven vulnerability scanner, in public beta. The feature is available immediately to Pro and Enterprise teams and, once enabled, automatically reviews every push to a GitHub repository connected to a Vercel project for issues such as missing authorization checks and exposed secrets.

How it works

Security Review is a new capability in Vercel Agent, sitting alongside Code Review and Investigation. It is triggered by GitHub push events rather than pull requests, so direct and force pushes to the production branch are reviewed as well. When a single push affects several projects, they are reviewed together, and in monorepos reviews are scoped to each project's root directory.

Each review covers only what changed, comparing a push against a base. For deployments, that base is the branch's previous deployment, meaning the change is reviewed against what is already running. Vercel says this keeps reviews fast enough to finish in seconds and run on every preview and production deployment.

Findings appear in the Vulnerabilities view of the Security dashboard. On private and internal repositories, Security Review also adds a GitHub check to the commit and posts findings as comments on the pull request; on public repositories, details stay in the dashboard. Each finding records the project, environment, branch and commit, along with a summary, a suggested fix and an AI-generated severity rating.

What the agent looks for

Vercel says the agent reads the diff together with its surrounding code, traces how the changed code can be reached, checks for existing protections, and rechecks each finding before reporting it. That process, the company argues, is what lets it catch logic flaws such as missing authorization checks, the kind of issue pattern-matching scanners tend to miss.

Teams can also give the agent context through security policies written in plain language, set either team-wide in Agent settings or per project in Security settings, and managed with the vercel security policy CLI command. A policy might, for example, require any endpoint returning customer data to verify that the requester owns it.

Fixing and tracking findings

Findings can be fixed in two ways. Clicking "Fix with Vercel Agent" hands the finding to Vercel Agent, which first confirms the vulnerability still exists, then makes the smallest change that fixes it and updates the tests. According to Vercel's changelog, the fix arrives as a pull request, and nothing merges without approval. Alternatively, teams can pull findings with vercel security vulnerabilities in the Vercel CLI (version 65.10.0 or later) and let their own coding agent handle the fix, or copy any finding as Markdown or JSON.

Findings are deduplicated against previously reported issues and tied to the deployment where they first appeared. If a vulnerable change from a pull request is merged into the production branch, the finding is flagged in production as well; when a later review no longer detects an issue in an environment, it is marked "no longer detected" and the GitHub comment is closed. Every new finding also generates an audit log event, which Enterprise teams can forward to their SIEM through Audit Log Drains.

Vercel's illustrative case: a developer pushes a change to an invoice download endpoint whose lookup does not check which organization the invoice belongs to. Security Review flags within seconds that any signed-in user could download another organization's invoice simply by changing its ID.

Pricing and limitations

Security Review is billed at the Vercel Agent rate: model tokens at the provider's underlying rate with no markup, plus a Vercel Token Rate of $0.25 per million tokens covering input, cached and output tokens. Because reviews run on every push, cost scales with how often a team pushes and how much code changes. Reviews are free until November 10, 2026; fixes made with Vercel Agent are billed at the rate above.

Vercel is explicit about the tool's limits: it analyzes source code, is not a security audit, and performs no runtime or penetration testing. It helps surface vulnerabilities but does not guarantee an application is free of them.

What's next

Security Review currently starts with code pushed to GitHub. Vercel says it is working on extending the feature to deployments from the Vercel CLI and v0, to repositories that do not deploy on Vercel at all, and on suggested fixes that can be applied directly from the pull request.

Why it matters

Vercel's argument for the feature is that frontier AI models are making vulnerabilities easier both to find and to exploit; the company points to researchers who last month turned two decade-old flaws in PHP core into working remote code execution exploits with relatively little effort. At the same time, agents are writing more of the code teams ship, and few apply consistent security review. Push-level, policy-aware scanning closes a real gap, since many review tools only examine pull requests and let direct pushes to production slip through unexamined. One caveat: both sources here are Vercel's own announcements, so effectiveness claims rest on the company's word until independent usage bears them out. Still, the launch signals where platform security tooling is heading: continuous, AI-driven review that treats every push, not every pull request, as the unit of scrutiny.