Journal

AI-built apps, safe to launch5 min read

Vercel Security Review now works without a linked project

Vercel's AI security check can now scan GitHub code that is not linked to a Vercel project. Here is what it catches, what it misses, and when you still need a person.

By

Researched and drafted with AI assistance, then checked and approved by Isaac Vicliph before it was published.

Since 13 November 2026, Vercel Security Review can scan any GitHub repository the Vercel GitHub App can access, even if the app is not deployed on Vercel. It is a paid, beta, code-only scanner. It is a useful first check, but it is not an audit or a penetration test, so an app holding personal information still needs a human review before launch.

What changed on 13 November

On 13 November 2026, Vercel said its Security Review can now scan GitHub repositories not linked to a Vercel project. Before this, the check only worked on code set up to deploy on Vercel. Now the code just needs to be on GitHub.

Security Review is an AI tool that reads your code and looks for security holes. A repository is the folder on GitHub that holds your app's code. By default, the new coverage includes every unlinked repository the Vercel GitHub App can access in the GitHub account you connect. Once it is turned on, every push (each time new code is sent up to GitHub) starts a review of the code that changed.

This was signposted. When Vercel launched Security Review on 2 November, it said repositories that do not deploy on Vercel were coming next. The 13 November change delivers that. The product is still described as a public beta, so expect it to keep changing.

Who can use it and what it costs

Security Review is only available to paid Pro and Enterprise teams. It is not on the free Hobby plan. The free period for reviews ended on 10 November 2026, and fixes made with Vercel Agent are billed too.

You pay per push, not per pull request. Vercel's Security Review documentation says cost grows with how often you push and how much code changed. If your AI coding tool pushes lots of small changes through the day, that adds up, so look at the pricing before you switch it on across every repository.

The code must sit in a GitHub repository the Vercel GitHub App can reach. Your app can be hosted somewhere else, but if the code only lives inside an AI builder tool, or on another Git host, this check does not cover it.

What an automated review is built to catch

Vercel says the tool is meant to find problems before they reach your users. It names three kinds of issue.

  • Code injection. where someone can slip their own commands into your app through a form, a web address or other input.
  • Authorisation bypass. where a user can reach data or actions they should not have, such as another customer's records.
  • Credential exposure. where passwords, API keys or other secrets end up somewhere they can be read.

What it does not check

Vercel's own documentation is clear on this. Security Review analyses your code, not your running app. It is not a security audit, and it does no runtime or penetration testing. Penetration testing means a person actively trying to break into the live app, the way an attacker would.

For each push, the AI agent reads the changed code and the code around it, traces how that code can be reached, checks for protections already in place, and rechecks each finding before reporting it. Each finding comes with a summary, a suggested fix and a severity rating generated by AI. Help Net Security described it as an AI code scanner that runs every time you push to GitHub. That is a fair way to think of it.

Unlinked repositories also get less to work with. When a push creates a deployment for a linked Vercel project, the review uses extra project and deployment context. An unlinked repository review does not have that. My own reading is that settings outside the code, such as database permissions, admin accounts and hosting configuration, sit outside what a code scanner can see. The AI severity rating is a starting point, not a final word.

When you still need a person to look at it

If your app stores personal information about real people, the stakes change. Under New Zealand's Privacy Act principle 5, organisations must have reasonable safeguards against loss, misuse or disclosure of personal information. An automated scan may be one part of your safeguards. Whether your safeguards are reasonable overall is a question for a privacy adviser or lawyer, not a scanner.

We would start with a human review before launch in these situations.

  • Personal or sensitive data. names, contact details, health, financial or identity information about customers or staff.
  • Logins and user roles. anything where different people should see different things, such as admins, staff and customers.
  • Payments. taking money or storing anything linked to payment.
  • Nobody has read the code. the app was built mostly by prompting an AI tool, and no developer has looked at how it fits together.

What to do about it this week

If you have an AI-built app heading towards real users, these steps are practical and quick. None of them need you to be a developer.

For context, this is the kind of work we did for StaffChecks Ltd, where we reviewed, secured and redesigned a new client portal before launch. It went live on 5 October 2026.

  • Find your code. confirm whether it is in a GitHub repository. If it only lives inside a builder tool, ask how to export it to GitHub.
  • Check your Vercel plan. Security Review needs a Pro or Enterprise team. On Hobby, it is not available.
  • Set a budget. because you pay per push, check how often your tools push code and decide which repositories to cover.
  • Read the first findings. work through the summaries and suggested fixes, starting with anything rated serious, and have someone check each fix.
  • List what it cannot see. write down your database settings, admin accounts and hosting setup, since a code scan does not test these.
  • Book a human review. if the app holds personal information, takes payments or has user roles, get a person to check it before launch.

Vercel Security Review is a sensible first filter for an AI-built app, and the 13 November change means more founders can use it. It reads your code, not your live app, so treat it as one check among several. If you want a person to review and secure your app before real users arrive, get your AI-built app checked.

Frequently asked questions

Is Vercel Security Review free?

No. Vercel offered free security reviews until 10 November 2026. Since then, reviews are billed, and fixes made with Vercel Agent are billed as well. You pay per Git push rather than per pull request, so the cost grows with how often you push code and how much changes. It is also only available to paid Pro and Enterprise teams, not the free Hobby plan.

Can Vercel Security Review check an app that is not hosted on Vercel?

Yes, if the code is on GitHub. Since 13 November 2026, Security Review can scan GitHub repositories that are not linked to a Vercel project. By default it covers all unlinked repositories the Vercel GitHub App can access in your connected GitHub account. Code that only lives inside an AI builder tool or on another Git host is not covered. Unlinked reviews also lack the extra project and deployment context linked projects get.

Is Vercel Security Review the same as a penetration test?

No. Vercel's documentation says Security Review analyses your code, not your running application. It is not a security audit and does not do runtime or penetration testing. Penetration testing means a person actively trying to break into the live app. Security Review is an AI code scanner that looks for issues such as code injection, authorisation bypass and credential exposure in the code you push.

Does an automated security scan meet my Privacy Act obligations in New Zealand?

A scan on its own should not be treated as meeting them. Privacy Act principle 5 says organisations must have safeguards that are reasonable in the circumstances to prevent loss, misuse or disclosure of personal information. An automated code scan can be part of that, but whether your safeguards are reasonable depends on your app and your data. Ask a privacy adviser or lawyer, and consider a human review before launch.

Sources

Share this article

  • #vercel
  • #app security
  • #ai-built apps
  • #github
  • #privacy act