Vercel launches security checks for every push in seconds
Vercel has launched Security Review: the service checks changed code on every GitHub push and scans for vulnerabilities, including missing authorization checks and exposed secrets, in seconds. Results appear in pull requests, the Security dashboard, and the CLI. An agent can verify a finding, make a minimal fix, and update tests, but the final decision remains with the team.
AI-processed from Vercel Blog; edited by Hamidun News
Security reviews often lose out to development speed. The code is already in a pull request, the build is green, and the release is waiting—while the manual audit happens sometime later. Vercel proposes embedding it directly into the familiar workflow: the new Security Review runs after every push to a connected GitHub repository and checks the changed code within seconds.
The service is available in public beta as part of Vercel Agent, alongside Code Review and Investigation. It looks not only for secrets in code but also for logical vulnerabilities—for example, a missing access-control check. Findings appear in the Security dashboard, GitHub Checks, and pull-request comments, and are also available through the CLI. With the team's permission, the agent can confirm the issue, prepare a minimal fix, and update the tests.
How it works in practice
Consider an endpoint for downloading an invoice. A developer retrieves a database record using only the invoice ID but does not check whether it belongs to the current user's organization. Technically, the user is authenticated, the request works, and the tests pass. However, supplying someone else's ID is enough to retrieve someone else's document.
Security Review reads the changed lines together with the surrounding code, traces the path to them, and checks whether the required protection already exists. In this example, it should detect that the query is not constrained by an organization ID. The proposed fix adds that constraint to the database query. After the next push, the service checks the change again and closes the previous finding if the issue is no longer detected.
This is an important distinction from simply searching for dangerous patterns. An authorization flaw rarely looks like a suspicious line: the vulnerability results from a missing check and becomes apparent only in the context of the route, user, and data model.
What the team gets
The review is triggered by GitHub push events, including direct pushes to the production branch. For a deployment, the previous deployed version serves as the baseline, so the agent compares the new code with what is already running. In a monorepo, the analysis scope is limited to the root directory of the relevant Vercel project.
The team can describe custom rules in natural language. For example: every endpoint that returns customer data must verify that the requested record belongs to the requesting user. Policies can apply to individual projects or the entire team and can also be managed through the Vercel CLI.
Findings are not recreated every time the same change moves between environments. Vercel links an issue to the deployment where it first appeared and tracks it from there. Enterprise teams can send audit events to a SIEM through Audit Log Drains. Details can also be exported as Markdown or JSON, so fixes do not have to be assigned to Vercel Agent: they can be passed to a team's own coding agent through the CLI.
Where the boundaries are
The main limitation is that the service analyzes only changes relative to the baseline version. This keeps reviews fast and makes it possible to run them on every push, but it does not turn the product into a full audit of the entire codebase. Older vulnerabilities unaffected by the new change may remain outside the current review. Even a second pass over findings does not eliminate false positives, manual review, or standard defensive layers: tests, static analysis, secret management, and dependency scanning.
For now, Security Review works with code pushed to GitHub. Vercel promises to add checks for deployments from the CLI and v0, as well as repositories that are not deployed on the Vercel platform.
The public beta is available to Pro and Enterprise teams. Reviews are free through November 10, 2026. After that, Vercel Agent pricing applies: model tokens are billed at the provider's price with no markup, plus $0.25 per million tokens. The practical next step for a team is simple: enable the review on one project first, define two or three rules for critical endpoints, and assess the quality of the findings before rolling it out across all repositories.
- Nov 12, 2026Claude in Chrome requires approval for every click on category 3 sites
- Nov 12, 2026Norwest raises record $4 billion fund, betting on agents, voice and physical AI
- Nov 12, 2026AWS shows how 100% retrieval precision can hide an AI agent's failure
- Nov 12, 2026Garage Eight dropped peer approvals for 69 MRs in an internal service using AI review
Key Questions Answered
How it works in practice?
Consider an endpoint for downloading an invoice. A developer retrieves a database record using only the invoice ID but does not check whether it belongs to the current user's organization. Technically, the user is authenticated, the request works, and the tests pass. However, supplying someone else's ID is enough to retrieve someone else's document. Security Review reads the changed lines together with the surrounding code, traces the path to them, and checks whether the required protection already exists.
What the team gets?
The review is triggered by GitHub push events, including direct pushes to the production branch. For a deployment, the previous deployed version serves as the baseline, so the agent compares the new code with what is already running. In a monorepo, the analysis scope is limited to the root directory of the relevant Vercel project. The team can describe custom rules in natural language. For example: every endpoint that returns customer data must verify that the requested record belongs to the requesting user.
Where the boundaries are?
The main limitation is that the service analyzes only changes relative to the baseline version. This keeps reviews fast and makes it possible to run them on every push, but it does not turn the product into a full audit of the entire codebase. Older vulnerabilities unaffected by the new change may remain outside the current review. Even a second pass over findings does not eliminate false positives, manual review, or standard defensive layers: tests, static analysis, secret management, and dependency scanning.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.