Skip to main content
Platform security guideReviewed

Vercel security for production, preview and branch deployments

A Vercel project serves the same code at several addresses, each with its own protection, variables and logs.

This guide explains the account and deployment settings that decide what each one exposes, what Vercel says is its job and what is yours, and what Vercel's terms say about testing. It comes with a worksheet that compares production, preview and branch deployments, control by control, using only read-only checks.

Vercel deployment exposure worksheet

Copy the table into a spreadsheet, one copy per project. For each control, read the setting where the second column points, write what you find under production, preview, and branch URLs and staging, and keep the evidence the last column lists.

How to use it

Every step reads a setting, list or log, or sends an ordinary request to one of your own deployment's URLs. Record that a value, token, bypass secret or link exists, its scope and who can read it, never the value itself. Plan conditions are as Vercel's documentation gave them on 14 November 2026.

13 rows

Vercel deployment exposure worksheet

ControlWhere to lookProductionPreviewBranch URLs and stagingEvidence to capture
Deployment Protection scope and methodProject Settings, then Deployment Protection. Team settings hold the default for new projects, which each project can override.Record the scope. "Standard Protection" leaves production domains public and restricts the production generated URL; "All Deployments" covers production domains too. Note whether a public production domain is intended.Record the method on preview URLs: Vercel Authentication on any plan, Password Protection as an add-on on Pro or included on the contract plan, or Trusted IPs or Passport on the contract plan.Record each Git branch URL in use. Under "(Legacy) Pre-Production Deployments", generated production branch URLs such as *-main.vercel.app stay reachable.A dated screenshot of the settings, plus a visit to each kind of URL in a private window not signed in to Vercel, noting whether the Vercel login or the site loaded.
Who can get past Vercel AuthenticationDeployment Protection, the Requests and Access sections, and the Share control on each deployment.If production is protected, list who holds access. Creating a Shareable Link for a production domain needs at least the Member team role or the Project Administrator role.List team members, project members, access groups and outside Vercel users granted access, plus pending requests. A Shareable Link for a non-production domain needs at least the Developer team role or the Project Developer role.List every Shareable Link by the branch it opens. On Hobby, one outside user and one Shareable Link in total per account.The Access list with names and dates, and Activity Log events for links created, regenerated or revoked. The link itself never goes in the worksheet, because holding it is the access.
Protection bypassesThe bypass sections of Deployment Protection, and Activity Log events project-automation-bypass, alias-protection-bypass-exception and project-options-allowlist.Each automation bypass secret works on every deployment in the project until revoked, production included where production is protected. Record which production webhooks rely on one.List each bypass secret by the name you gave it and the tool that uses it, and whether it is sent as a header or inside a URL held by another service.List each domain in "Deployment Protection Exceptions", each "OPTIONS Allowlist" path, and each Trusted Source with the environments it may reach.Names, owners, places of use and dates only. Record that a secret exists, who can manage it and where it is used, never its value.
Environment variable scope and typeEnvironment Variables in the project sidebar, and the team-wide list in Team Settings. Filter by Environment; each variable shows a Config or Secret label.List each key applied to Production with its label, and whether the "Require Separate Values" policy, which users with the owner role can enable, is on.List each key applied to Preview and whether it covers all non-production branches or one branch. Flag any that hold a live payment, email or database credential, since a Preview value for all non-production branches reaches each of their builds unless a branch value overrides it.List branch-specific overrides, custom environment variables and variables added by an integration. Check vercel.json for legacy env or build.env properties.Key names, labels, environments and branches. Never a value, and no CLI command that decrypts one.
Who can read or change variable valuesThe team's member list with each person's role, and the Activity Log filtered to environment variable events.List who can change Production values. The Developer role cannot alter production environment variables.List who can change Preview and Development values, which the Developer role can. Config values stay readable to anyone whose role gives access to the project's variables.List who can create or change custom environments: team owners and project admins.A dated role list, plus env-variable-read events and the CLI decryption events for env pull, env ls and pull, with names and dates.
What starts a deployment, and with whose codeProject Settings, the Security section for Git Fork Protection, and the Environments settings for branch tracking and "Auto-assign Custom Production Domains".Record the production branch and who can merge to it, since the Developer role can deploy to production by merging. Record whether production domains are assigned automatically or wait for promotion.Record whether Git Fork Protection is on. On GitHub, with it on, Vercel requires authorisation from you or a team member before deploying a pull request from a fork.Record each tracked branch, the environment it deploys to, and whether that environment uses staging or production credentials.Dated screenshots of the Git, Security and Environments settings.
Function logs and drainsLogs in the project sidebar, filtered by Environment and Branch. Build logs open from the production deployment tile.Record your plan's runtime log retention: 1 hour on Hobby, 1 day on Pro, 3 days on the contract plan, 30 days on Pro or the contract plan with Observability Plus. Record any drain and where it sends logs.Runtime logs cover preview deployments too. Open one sampled request and note whether its log lines contain keys, tokens or personal details.Filter by each tracked branch and repeat the sample. Note whether any function logs request bodies or headers.Retention, drain names and destinations, and for each sample the route and field name of anything sensitive, never the value.
Custom domains and generated URLsThe project's Domains settings, and the deployment overview, which can list generated URLs beside custom domains.List each production domain and the production generated URL, and record which of them the protection scope covers.Record the unique deployment URLs and, for CLI deployments on a team, author URLs. Vercel adds an X-Robots-Tag: noindex header to preview deployments.List each custom domain assigned to a Git branch. Vercel does not add the noindex header there, so the domain can be indexed unless protected or given the header.The domain list with the environment and branch each serves, and the response headers of each kind of URL, read with curl -I.
Old deployments that still answerProject Settings, Build and Deployment, Deployment Retention Policy, and the team default in team settings.Record the retention for production deployments. An older production build keeps the code and variable values it was built with for as long as it is kept.Record the retention for pre-production deployments. Unless a custom policy is set, the default is 30 days on Hobby and Pro and 180 days on the contract plan.Note deployments kept regardless of policy. They include the latest Ready deployment of a Git branch with a domain assigned, any non-production deployment with a custom alias, and the latest preview of a branch that has not been deleted and has an open pull request.A dated screenshot of the policy, and Activity Log events project-deployment-retention-updated.
Firewall rules and bypassesFirewall in the project sidebar, its rules list, and the View Audit Log option on that page.List each custom rule with its action and the hosts or paths in its conditions, each IP block, and each managed ruleset with its mode. Publishing applies changes to the production deployment.Record whether any rule's conditions name preview hosts. Where nothing says a rule covers preview URLs, write "not confirmed".List each System Bypass Rule and each custom rule with a bypass action, with the reason and who added it, including addresses set to Bypass for a past test.Screenshots of the rules list, the plan limits that apply, and Activity Log events firewall-config-modified and firewall-bypass-created.
Framework and runtime advisoriesThe lockfile of the commit each environment was built from, the framework's own advisory list, and Vercel's Security Bulletins page.Record the framework and its resolved version in the deployment serving production, and any bulletin that names it.Record the same for each long-lived preview, which runs the version it was built with until it is rebuilt.Record the same for staging and for any older deployment still inside its retention period.Version, commit and date per environment, and each advisory list or bulletin checked with the date read. For Next.js, compare against the Next.js guide's advisory ledger.
Team access, two-factor sign-in and historyThe team's member list; the team's two-factor requirement; the Activity Log, or Audit Logs on the contract plan.List owners, and members, whom Vercel describes as having full control over projects. Vercel recommends at least two people with the owner role.List developers and viewers. Any logged-in team member with at least a viewer role can open previews behind Vercel Authentication.List outside Vercel users granted access, and Contributors with project roles on the contract plan.Whether owners require two-factor sign-in, the dated member list, and on the contract plan an Audit Log CSV export for the review period, which reaches back up to 90 days.
Plan and testing terms before any active testVercel's "Penetration testing on Vercel" guide, the Fair Use Guidelines, the Acceptable Use Policy and your plan.Record the plan. Vercel's guide says penetration testing is not permitted for Hobby users, and that Pro and contract plan customers may test their own applications but not the underlying infrastructure and platform.Staged production deployments use production variables, so testing there can reach production services and data. Record which environment a test would target and whose credentials it carries.Vercel's environments page points to a custom environment or preview branch with staging credentials when separate resources are needed. List the downstream suppliers the target environment calls, since Vercel's guide asks for consent from all relevant parties.The date each policy page was read, the plan, supplier consents, and the owner's signed authorisation. Volumetric testing is limited to the contract plan with advance notice to Vercel.

Control

Where to look

  1. Deployment Protection scope and method
    Where to look
    Project Settings, then Deployment Protection. Team settings hold the default for new projects, which each project can override.
    Production
    Record the scope. "Standard Protection" leaves production domains public and restricts the production generated URL; "All Deployments" covers production domains too. Note whether a public production domain is intended.
    Preview
    Record the method on preview URLs: Vercel Authentication on any plan, Password Protection as an add-on on Pro or included on the contract plan, or Trusted IPs or Passport on the contract plan.
    Branch URLs and staging
    Record each Git branch URL in use. Under "(Legacy) Pre-Production Deployments", generated production branch URLs such as *-main.vercel.app stay reachable.
    Evidence to capture
    A dated screenshot of the settings, plus a visit to each kind of URL in a private window not signed in to Vercel, noting whether the Vercel login or the site loaded.
  2. Who can get past Vercel Authentication
    Where to look
    Deployment Protection, the Requests and Access sections, and the Share control on each deployment.
    Production
    If production is protected, list who holds access. Creating a Shareable Link for a production domain needs at least the Member team role or the Project Administrator role.
    Preview
    List team members, project members, access groups and outside Vercel users granted access, plus pending requests. A Shareable Link for a non-production domain needs at least the Developer team role or the Project Developer role.
    Branch URLs and staging
    List every Shareable Link by the branch it opens. On Hobby, one outside user and one Shareable Link in total per account.
    Evidence to capture
    The Access list with names and dates, and Activity Log events for links created, regenerated or revoked. The link itself never goes in the worksheet, because holding it is the access.
  3. Protection bypasses
    Where to look
    The bypass sections of Deployment Protection, and Activity Log events project-automation-bypass, alias-protection-bypass-exception and project-options-allowlist.
    Production
    Each automation bypass secret works on every deployment in the project until revoked, production included where production is protected. Record which production webhooks rely on one.
    Preview
    List each bypass secret by the name you gave it and the tool that uses it, and whether it is sent as a header or inside a URL held by another service.
    Branch URLs and staging
    List each domain in "Deployment Protection Exceptions", each "OPTIONS Allowlist" path, and each Trusted Source with the environments it may reach.
    Evidence to capture
    Names, owners, places of use and dates only. Record that a secret exists, who can manage it and where it is used, never its value.
  4. Environment variable scope and type
    Where to look
    Environment Variables in the project sidebar, and the team-wide list in Team Settings. Filter by Environment; each variable shows a Config or Secret label.
    Production
    List each key applied to Production with its label, and whether the "Require Separate Values" policy, which users with the owner role can enable, is on.
    Preview
    List each key applied to Preview and whether it covers all non-production branches or one branch. Flag any that hold a live payment, email or database credential, since a Preview value for all non-production branches reaches each of their builds unless a branch value overrides it.
    Branch URLs and staging
    List branch-specific overrides, custom environment variables and variables added by an integration. Check vercel.json for legacy env or build.env properties.
    Evidence to capture
    Key names, labels, environments and branches. Never a value, and no CLI command that decrypts one.
  5. Who can read or change variable values
    Where to look
    The team's member list with each person's role, and the Activity Log filtered to environment variable events.
    Production
    List who can change Production values. The Developer role cannot alter production environment variables.
    Preview
    List who can change Preview and Development values, which the Developer role can. Config values stay readable to anyone whose role gives access to the project's variables.
    Branch URLs and staging
    List who can create or change custom environments: team owners and project admins.
    Evidence to capture
    A dated role list, plus env-variable-read events and the CLI decryption events for env pull, env ls and pull, with names and dates.
  6. What starts a deployment, and with whose code
    Where to look
    Project Settings, the Security section for Git Fork Protection, and the Environments settings for branch tracking and "Auto-assign Custom Production Domains".
    Production
    Record the production branch and who can merge to it, since the Developer role can deploy to production by merging. Record whether production domains are assigned automatically or wait for promotion.
    Preview
    Record whether Git Fork Protection is on. On GitHub, with it on, Vercel requires authorisation from you or a team member before deploying a pull request from a fork.
    Branch URLs and staging
    Record each tracked branch, the environment it deploys to, and whether that environment uses staging or production credentials.
    Evidence to capture
    Dated screenshots of the Git, Security and Environments settings.
  7. Function logs and drains
    Where to look
    Logs in the project sidebar, filtered by Environment and Branch. Build logs open from the production deployment tile.
    Production
    Record your plan's runtime log retention: 1 hour on Hobby, 1 day on Pro, 3 days on the contract plan, 30 days on Pro or the contract plan with Observability Plus. Record any drain and where it sends logs.
    Preview
    Runtime logs cover preview deployments too. Open one sampled request and note whether its log lines contain keys, tokens or personal details.
    Branch URLs and staging
    Filter by each tracked branch and repeat the sample. Note whether any function logs request bodies or headers.
    Evidence to capture
    Retention, drain names and destinations, and for each sample the route and field name of anything sensitive, never the value.
  8. Custom domains and generated URLs
    Where to look
    The project's Domains settings, and the deployment overview, which can list generated URLs beside custom domains.
    Production
    List each production domain and the production generated URL, and record which of them the protection scope covers.
    Preview
    Record the unique deployment URLs and, for CLI deployments on a team, author URLs. Vercel adds an X-Robots-Tag: noindex header to preview deployments.
    Branch URLs and staging
    List each custom domain assigned to a Git branch. Vercel does not add the noindex header there, so the domain can be indexed unless protected or given the header.
    Evidence to capture
    The domain list with the environment and branch each serves, and the response headers of each kind of URL, read with curl -I.
  9. Old deployments that still answer
    Where to look
    Project Settings, Build and Deployment, Deployment Retention Policy, and the team default in team settings.
    Production
    Record the retention for production deployments. An older production build keeps the code and variable values it was built with for as long as it is kept.
    Preview
    Record the retention for pre-production deployments. Unless a custom policy is set, the default is 30 days on Hobby and Pro and 180 days on the contract plan.
    Branch URLs and staging
    Note deployments kept regardless of policy. They include the latest Ready deployment of a Git branch with a domain assigned, any non-production deployment with a custom alias, and the latest preview of a branch that has not been deleted and has an open pull request.
    Evidence to capture
    A dated screenshot of the policy, and Activity Log events project-deployment-retention-updated.
  10. Firewall rules and bypasses
    Where to look
    Firewall in the project sidebar, its rules list, and the View Audit Log option on that page.
    Production
    List each custom rule with its action and the hosts or paths in its conditions, each IP block, and each managed ruleset with its mode. Publishing applies changes to the production deployment.
    Preview
    Record whether any rule's conditions name preview hosts. Where nothing says a rule covers preview URLs, write "not confirmed".
    Branch URLs and staging
    List each System Bypass Rule and each custom rule with a bypass action, with the reason and who added it, including addresses set to Bypass for a past test.
    Evidence to capture
    Screenshots of the rules list, the plan limits that apply, and Activity Log events firewall-config-modified and firewall-bypass-created.
  11. Framework and runtime advisories
    Where to look
    The lockfile of the commit each environment was built from, the framework's own advisory list, and Vercel's Security Bulletins page.
    Production
    Record the framework and its resolved version in the deployment serving production, and any bulletin that names it.
    Preview
    Record the same for each long-lived preview, which runs the version it was built with until it is rebuilt.
    Branch URLs and staging
    Record the same for staging and for any older deployment still inside its retention period.
    Evidence to capture
    Version, commit and date per environment, and each advisory list or bulletin checked with the date read. For Next.js, compare against the Next.js guide's advisory ledger.
  12. Team access, two-factor sign-in and history
    Where to look
    The team's member list; the team's two-factor requirement; the Activity Log, or Audit Logs on the contract plan.
    Production
    List owners, and members, whom Vercel describes as having full control over projects. Vercel recommends at least two people with the owner role.
    Preview
    List developers and viewers. Any logged-in team member with at least a viewer role can open previews behind Vercel Authentication.
    Branch URLs and staging
    List outside Vercel users granted access, and Contributors with project roles on the contract plan.
    Evidence to capture
    Whether owners require two-factor sign-in, the dated member list, and on the contract plan an Audit Log CSV export for the review period, which reaches back up to 90 days.
  13. Plan and testing terms before any active test
    Where to look
    Vercel's "Penetration testing on Vercel" guide, the Fair Use Guidelines, the Acceptable Use Policy and your plan.
    Production
    Record the plan. Vercel's guide says penetration testing is not permitted for Hobby users, and that Pro and contract plan customers may test their own applications but not the underlying infrastructure and platform.
    Preview
    Staged production deployments use production variables, so testing there can reach production services and data. Record which environment a test would target and whose credentials it carries.
    Branch URLs and staging
    Vercel's environments page points to a custom environment or preview branch with staging credentials when separate resources are needed. List the downstream suppliers the target environment calls, since Vercel's guide asks for consent from all relevant parties.
    Evidence to capture
    The date each policy page was read, the plan, supplier consents, and the owner's signed authorisation. Volumetric testing is limited to the contract plan with advance notice to Vercel.

Where Vercel sits in your application

Vercel security starts with knowing which parts of the stack Vercel runs for you.

Vercel builds each deployment, serves it at its own URLs, runs your server code in a compute environment that its shared responsibility model says uses Vercel Functions and containers, and puts a firewall in front of every request. By default it creates a preview deployment when a commit is pushed to a branch that is not the production branch, when a pull request is created, and when the CLI deploys without the --prod flag. Those previews sit beside the production deployment that answers on your domain.

That shape is why this guide compares environments instead of listing settings. A production site can be locked down while an old preview, a staging branch on its own subdomain or a build carrying production keys sits next to it under different rules. The worksheet on this page puts production, preview and branch deployments side by side for each control, so a gap shows up as a difference between columns.

Plan matters throughout, because Vercel attaches several controls to a plan. Its plans page lists Hobby, which is free, then Pro, then a plan that includes all features of the Pro plan with custom limits. This page calls the plan listed after Pro the contract plan.

Section 18.1 of Vercel's Terms of Service sets that plan apart from the free hobby plan and the paid self-service subscription plans: for it, a customer may contact Vercel separately.

Each plan condition on this page is the one Vercel's documentation gave when this page was reviewed on 14 November 2026. Check Vercel's page for a feature before you rely on its plan.

This guide covers account and deployment settings. What your code does with a request, such as Next.js middleware, Server Actions and caching, is the subject of the Next.js security guide, which also keeps the framework advisory ledger.

What Vercel secures and what stays with you

Asking whether Vercel is secure is really asking how the duties are split.

Vercel's shared responsibility model says the customer handles their data, applications and user access management, while Vercel manages infrastructure components such as compute, storage and networking. Vercel is responsible for detecting, investigating and remediating security incidents that affect the infrastructure and services it manages.

Customers monitor and respond to incidents in their own application code, configuration and integrations. Vercel's security overview adds that every deployment is served over HTTPS, with certificates Vercel generates for the unique deployment URLs.

Five items in that model bear directly on the settings in this guide:

  • Code and environment variables: the customer manages the application's code, including the exposure of environment variables, and the handling of secrets such as API keys.
  • Authentication: the customer handles the application's own sign-in. Vercel manages sign-in to the platform and provides Deployment Protection.
  • Logs: Vercel provides short-term runtime logs, and long-term retention through log drains is the customer's to set up.
  • Malicious traffic: the customer carries the costs and resource use that malicious traffic causes, and should assess its security requirements and add safeguards beyond the protections Vercel provides.
  • Fit: the customer decides whether Vercel's platform and the protection it provides meet the application's needs.

A duty written into the terms

Section 10.1 of Vercel's Terms of Service says the customer shall configure their content, including projects and deployments, so that its transmission, storage or use will not expose personal data or personal information without proper consent from individuals as determined by applicable law.

What that means for your business is a question for your own adviser. For this guide, it is a reminder that every row of the worksheet is a setting the customer controls.

Deployment Protection: what each scope leaves public

Deployment Protection decides who can open a deployment's URL.

It is set per project, with a team default for new projects that a project can override, by choosing a scope (which URLs are covered) and a method (how a visitor proves they may enter). Vercel says it requires authentication for all requests, including those to Routing Middleware.

The scope is the setting that decides what stays public. "Standard Protection" covers every deployment except production domains, is available on all plans, and is the option Vercel recommends for most projects. "All Deployments" covers every URL including production domains, also on all plans. The settings also list legacy scopes.

"(Legacy) Standard Protection" leaves up-to-date production URLs unprotected, and "(Legacy) Pre-Production Deployments" covers only preview URLs, so past and present generated production branch URLs such as *-main.vercel.app stay reachable. Protecting production alone uses Trusted IPs, is a contract plan option, and leaves preview URLs public.

The method depends on plan. Vercel Authentication is available on every plan. Password Protection is not available on Hobby, is a paid add-on on Pro and is included on the contract plan. Trusted IPs and Passport, which lets visitors sign in through your own identity provider, are contract plan features. Protected Source Maps puts requests for .map files behind Vercel Authentication.

Two Vercel pages describe the starting point differently. The environments page says new projects start with Standard Protection and Vercel Authentication unless the team has set another default, which keeps preview URLs behind a Vercel login and leaves production domains public.

The generated URLs page says each deployment's unique URL is publicly accessible by default and can be made private with deployment protection. Read the project's own setting rather than either default. It is the first row of the worksheet.

Who gets past Vercel Authentication

With Vercel Authentication on, a deployment opens for logged-in team members with at least a viewer role, project members with at least the project Viewer role, members of an access group that can reach the project, Vercel users who have been granted access, anyone holding a Shareable Link, and tools that send the automation bypass header.

People outside the team can request access, and those requests can be approved or declined in the Requests section of the project's Deployment Protection settings or in the share modal on the deployment page. On Hobby, one external user per account can be granted access.

Three behaviours matter when the setting changes or someone signs in. Turning Vercel Authentication off leaves every existing deployment unprotected. Turning it back on lets a user who had already signed in to a deployment, and still has its cookie in their browser, keep access without signing in again.

The cookie itself works for one URL only, even between URLs that point to the same deployment. Signing in on a branch URL therefore does not set the cookie for that deployment's unique URL.

Every way around Deployment Protection needs a named owner

Vercel's page on bypassing Deployment Protection lists these methods, each available on every plan: Trusted Sources, Protection Bypass for Automation, OPTIONS Allowlist, Deployment Protection Exceptions and Shareable Links.

Each has a sound use. Each also lets some requests reach a protected deployment without the caller signing in to Vercel. Keep each in an inventory with a named owner.

  • Protection Bypass for Automation: a project can hold several bypass secrets, each works on every deployment in the project until it is revoked, and Vercel sets one of them as the VERCEL_AUTOMATION_BYPASS_SECRET system environment variable. A valid token skips Vercel Authentication, Password Protection and Trusted IPs, gets past the firewall's system mitigations and does not trigger bot protection challenges. It does not override active DDoS mitigations, or the rate limits and challenges applied during a detected attack.
  • How the bypass secret travels: as the x-vercel-protection-bypass header, which Vercel recommends for testing tools, or as a query parameter, which Vercel says is required for webhook URLs from services that cannot set headers.
  • Shareable Links: a link with a query parameter that opens a branch deployment to anyone holding it. Developers on Hobby can create one in total per account.
  • Deployment Protection Exceptions: a preview domain added here becomes publicly accessible, bypassing protection features such as Vercel Authentication, Password Protection and Trusted IPs.
  • OPTIONS Allowlist: OPTIONS requests to any path that starts with a listed path skip protection, so listing /api leaves every path starting with /api unprotected for OPTIONS requests.
  • Trusted Sources: other Vercel projects or outside services such as GitHub Actions present a short-lived OIDC token, and access is limited to matching environments by default.

A secret in a webhook address lives somewhere else

A bypass secret written into a webhook URL is stored in the settings of the service that sends the webhook, which your team does not control. Vercel's own guidance is to keep the secret in an environment variable rather than writing it into the webhook URL configuration.

Vercel also notes that regenerating or deleting a secret invalidates previous deployments, and that the app needs a redeploy to use the new value. Rotation is therefore planned work, and it sits outside this read-only worksheet. The worksheet records that each secret exists, who can manage it and where it is used, and never its value.

Vercel lists the people who can manage these secrets as team members with at least the member role and project members with the Project Administrator role.

Environment variables: scope, type and who can read them

Each variable applies to one or more environments: Production, Preview, Development and, on Pro and the contract plan, custom environments such as staging.

A Preview variable applies either to every branch that is not the production branch or to one branch you select, and a branch-specific value overrides other Preview values with the same name. Values are encrypted at rest. A Config value stays readable to members whose role gives access to the project's environment variables, while a Secret is write-only once saved.

A change applies only to new deployments, so an older deployment keeps the values it was built with.

Three cases hand production values to builds that people think of as previews. The first deployment of a new project is always a production deployment and builds with Production variables, whichever branch it came from.

A staged production deployment, made by turning off "Auto-assign Custom Production Domains", uses production variables, and Vercel's environments page warns that testing it can reach production services and data. Promoting a preview to production rebuilds it with Production variables.

The opposite mistake is a live credential added to Preview for all non-production branches, which then reaches each of their builds unless a branch value overrides it.

Settings that narrow who sees a value

Owners can turn on the "Require Separate Values" team policy, after which a Secret cannot share its Production value with Preview, Development or a custom environment. The Developer role cannot alter production environment variables, although developers can deploy to production by merging to the production branch. A variable an integration added is labelled with that integration in project settings and cannot be switched to Secret from the list.

If vercel.json still holds env or build.env properties, Vercel's documentation calls them legacy and shows how to move them into the Environment Variables settings. Anything in that file travels with your code.

The Activity Log, available on every plan, records an env-variable-read event when the plain text value of an encrypted variable is read, and separate events when values are decrypted through CLI commands such as env pull and env ls.

That gives the worksheet its evidence of who has read values, without anyone opening one. On GitHub, pull requests from forks are covered by Git Fork Protection: Vercel requires authorisation from you or a team member before deploying them, which Vercel says protects against leaking sensitive project information such as environment variables and the OIDC token.

It can be turned off in the Security section of Project Settings. Read its state rather than assume it.

One framework detail crosses into code: Next.js inlines NEXT_PUBLIC_ variables into the client bundle at build time. Treat any NEXT_PUBLIC_ value as public, whatever label it carries in Vercel. The Next.js guide covers that, and the secrets exposure guide covers what to do once a key has leaked.

Vercel firewall and WAF: what runs by default and what you set

Every incoming request passes through the platform-wide firewall and the Vercel WAF.

The platform-wide firewall, with DDoS mitigation, is free on every plan and needs no configuration. The Vercel WAF, also available on every plan, is where you add custom rules, IP blocks, managed rulesets and Attack Mode. Rules run in a fixed order: DDoS mitigation, then WAF IP blocking, then custom rules, then managed rulesets.

The plan sets how far the WAF goes. Per project, Hobby allows up to 3 custom rules and 10 IP blocks, Pro up to 40 of each, and the contract plan up to 250 of each. System Bypass Rules are not available on Hobby, with up to 10 on Pro and up to 75 on the contract plan.

The OWASP core ruleset is a contract plan feature that Hobby and Pro do not have. Persistent actions, which block a client's IP address for a set time after a rule challenges or denies it, are on Pro and the contract plan.

The Bot Protection Managed Ruleset and the AI Bots Managed Ruleset are on every plan but inactive by default, shown as "Off" and "Allow" respectively. In both cases matching traffic is not evaluated and reaches the application. Custom rules run before managed rulesets, and a custom rule with a bypass action can let through traffic a managed ruleset would block. Give bypass rules the same review as bypass secrets.

How firewall changes take effect

A configuration change takes effect globally within 300ms, needs no redeploy, and can be rolled back from the View Audit Log option on the project's Firewall page. Vercel's recommended way to add a rule is to start with a log action, watch the 10-minute live traffic view, and only then switch to challenge, deny, bypass or rate limit. Deny or challenge rules can also catch the requests Vercel sends to start your Cron Jobs.

Vercel's documentation says publishing applies firewall changes to the production deployment, and the Firewall overview graphs traffic for the production deployment. Work out what a rule does on preview and branch hosts from its conditions, and write "not confirmed" in the worksheet rather than assume. Choosing, tuning and testing WAF rules in general is the subject of the web application firewall guide; this page stays with what Vercel's settings say.

Function logs, retention and drains

Runtime logs hold the output of Vercel Functions in both preview and production deployments, including console.log output, and can be filtered by environment, branch, host and deployment.

How long they last depends on plan: 1 hour on Hobby, 1 day on Pro and 3 days on the contract plan, or 30 days on Pro or the contract plan with Observability Plus, which lets you view up to 14 consecutive days of runtime logs within that 30-day window.

Vercel's runtime logs page says runtime logs are viewed from the Logs tab, and points to the production deployment tile for build logs.

On Hobby, then, a runtime log entry can no longer be read an hour after it was written. Vercel's shared responsibility model puts long-term retention on the customer, through log drains. Drains, on Pro and the contract plan, forward runtime, build and static logs to an outside service, and Vercel documents signature checks for securing them. Audit Log Drains are a contract plan feature.

Vercel documents redaction in build logs: during builds, a Secret value of 32 characters or longer that appears in a build log is replaced, and the automation bypass secret and the OIDC token are always redacted there. What a function writes at runtime is your code's output, so the worksheet asks you to read one sampled request in each environment for keys, tokens and personal details, and to record any you find by field name only.

Custom domains, generated URLs and old deployments

One deployment can answer at several addresses, including its unique deployment URL, a Git branch URL of the form project-git-branch-scope.vercel.app, the production URL, any custom domain, and, for CLI deployments on a team, a URL tied to the deployment's author.

Because the protection scopes are defined by kinds of URL, each address has to be checked against the scope rather than assumed to share the production domain's settings.

Vercel adds an X-Robots-Tag: noindex header to preview deployments and to production deployments opened through their unique URL. It does not add the header when a preview is served from a custom domain assigned to a Git branch, such as a staging subdomain, and search engines can index that domain unless it is protected or the header is added.

Vercel suggests checking a URL by running curl -I against it and looking for the x-robots-tag header in the response; do that for each kind of URL, and read the other security headers your app sends while you are there.

Generated URLs keep working for as long as the deployment retention policy allows. Unless a team or project uses a custom policy, the default retention on Hobby and Pro is 30 days for cancelled, errored, pre-production and production deployments alike. On the contract plan the defaults are 30 days for cancelled, 90 days for errored, 180 days for pre-production and 1 year for production deployments.

Some deployments are kept whatever the policy says, including any with a production alias, the latest Ready deployment for a Git branch that has a domain assigned, and the latest preview deployment for a branch that has not been deleted and has an open pull request. Each of those runs the code and variable values it was built with.

When a custom domain leaves a project, the DNS record that pointed it at Vercel is the next thing to check, and the subdomain takeover guide explains why.

Team roles, two-factor sign-in and account history

Team roles are available on Pro and the contract plan.

Vercel lists them as Owner, Member, Developer, Security, Billing, the viewer roles and Contributor, and its access roles page gives the Security and Contributor roles as available on the contract plan. Only owners can change roles, and Vercel recommends that at least two people hold the owner role for continuity.

Members can administer security settings for the projects assigned to them. The Security role manages the firewall and IP blocking and cannot create deployments by default.

Any logged-in team member with at least a viewer role can open deployments that sit behind Vercel Authentication. The member list therefore belongs in any review of who can see protected previews.

Team owners can require two-factor authentication before members reach team resources, and a member turning on personal two-factor sign-in does not switch on that team requirement. Vercel supports authenticator apps and passkeys, and says passkeys protect against phishing by checking the site's identity.

The Activity Log and Audit Logs show who changed what. The Activity Log, on every plan and visible to the account owner and team members, records protection, bypass, firewall, drain, retention and environment variable events with the people involved and the time.

Audit Logs, on the contract plan and open to the owner and security roles, add the IP address and user agent behind each action and can be exported as a CSV file for the last 90 days.

Framework advisories and Vercel security bulletins

A search for a Vercel vulnerability can turn up two different things: a flaw in one of Vercel's own products, or a flaw in a framework or runtime your deployments run.

Vercel's Security Bulletins page publishes security updates and advisories for Vercel products. On 14 November 2026, the summaries that page gave for bulletins on flaws in Node.js, PHP and Better Auth said that Vercel WAF protections or blocks of known exploit patterns do not replace patching. Check that page for anything published since.

That makes the deployed version the evidence. Each deployment runs the code it was built from, so production, a long-lived preview and an old deployment still inside its retention period can carry different framework versions. The worksheet records the resolved version from the lockfile of the commit each environment was built from, and the advisory list you checked it against.

For Next.js applications, the advisory ledger on the Next.js guide lists affected and fixed releases with the conditions to compare.

How to fill in the deployment exposure worksheet

Make one copy per project and work row by row with the project open in the dashboard.

Fill each environment column from what the setting shows for that environment. Where a control does not exist on your plan, write "not available" and the plan rather than leaving the cell blank, because a blank reads as unchecked.

  • Every check reads a setting, a list or a log, or sends an ordinary request to one of your own deployment's URLs. None changes a setting, creates or revokes a link or secret, or decrypts a value.
  • Record names, labels, scopes, environments, people and dates. Never copy an environment variable value, a token, a bypass secret or a Shareable Link into the worksheet. Write that it exists, its scope and who can read or use it.
  • Read variable names and their Config or Secret labels from the dashboard list, filtered by environment. Vercel's Activity Log records the CLI commands env pull and env ls as decryption events. Leave both out of a review.
  • For URL rows, open your own production, preview and branch addresses in a private browser window that is not signed in to Vercel and note what loads. That is one ordinary page request to your own deployment.
  • Date every screenshot, and remove the names of people outside your business before the worksheet is shared beyond the team.
  • Repeat the review after a plan change, a new team member, a new integration or a change of protection scope.

What a review of your own account cannot show

The worksheet records configuration.

It shows which URLs sit behind a login, which builds receive which variables and which rules the firewall holds. It does not show what your application does with a request that reaches it: whether an API route checks who owns a record, whether a function returns more data than the screen needs, or whether a page sends a key to the browser in its JavaScript. Those are questions for a test of the application itself.

It cannot see Vercel's side of the shared responsibility model either. Vercel says it performs continual security assurance against its underlying platform, including penetration testing, and makes recent pentest reports available to Pro and contract plan customers at security.vercel.com. Evidence about the platform comes from there, not from a review of your settings.

It is also a snapshot. Logs reach back only as far as your plan's retention, a setting can change the day after you read it, and the Activity Log is the record of what changed in between.

Vercel's testing terms, read before any active test

Vercel's guide "Penetration testing on Vercel" says penetration testing is authorised for Pro and contract plan customers and is not permitted for Hobby users, whom it refers to the Fair Use Guidelines.

Pro and contract plan customers may test their own applications hosted on Vercel and must not target the underlying hosting infrastructure and platform. Non-volumetric testing on Pro and the contract plan needs no advance notice to Vercel.

Volumetric testing, such as sending a large number of requests from an automated scanner, is limited to the contract plan, with advance notice to Vercel through the customer's Customer Success Manager or Account Executive.

The same guide asks customers to consider how testing could affect downstream suppliers, such as backend or database hosts, and to obtain consent from all relevant parties. It says the customer pays all variable costs of the testing, and it recommends setting the testers' IP addresses to Bypass in the project's WAF configuration so other rules do not block them.

That last step changes the firewall, so it is a decision for the account owner, recorded in the worksheet with the addresses and dates.

Vercel's Fair Use Guidelines, a separate page, say Vercel usage is subject to them across plans and usage-based resources, and give a rule of thumb for which projects fall within Vercel's definition of fair use. Penetration testing is in the list headed "Never fair use", which names no plan.

For further information the guidelines point to the Terms of Service or, for the contract plan, its service agreement. Vercel's Acceptable Use Policy, which section 8 of the Terms incorporates by reference, prohibits creating an undue burden on Vercel's or third parties' websites, infrastructure, services or networks, undermining the security or integrity of the computing systems or networks of Vercel, its partners or any other person, and attempting to gain unauthorised access to the Services or related systems or networks.

Read the guide, the Fair Use Guidelines and the Acceptable Use Policy as they stand on the day, record the date, and where they leave doubt for your plan, get Vercel's answer in writing before any active test is booked.

Where a Clever Ops assessment fits

This guide is informational.

Clever Ops does not test Vercel's platform. The Clever Ops methodology puts a host's platform outside the method, and Vercel's guide says customers must not target its underlying hosting infrastructure and platform.

Your own application is a different matter. Clever Ops runs a web application and API penetration test on custom web applications and their APIs, such as applications built on Next.js, React or Node, with test accounts in each role the scope names.

Active testing of an application hosted on Vercel sits within that scoped assessment. Before it begins, each target's owner signs an authorisation and Clever Ops checks the host's testing policy. While it runs, the request rate is held to an agreed limit, and testing on a target pauses if a firewall starts blocking. Clever Ops refers load and denial-of-service testing to a load-testing specialist, which needs your host's written approval.

Before testing, Clever Ops asks for the exact targets, such as domains, subdomains and API addresses, plus anything that must be left alone. A completed worksheet already lists those addresses for each environment, along with the access routes and which environment carries which credentials.

Work on an application Clever Ops built or hosts is reported as a security review, not an independent penetration test, and its report names the builder and the host.

Round-the-clock monitoring and emergency incident response are not Clever Ops services. If you think an account or key has already been misused, contact Vercel and an incident response specialist first.

Frequently asked questions

Vercel answers with a shared responsibility model rather than a single verdict. It says Vercel manages infrastructure components such as compute, storage and networking, and the customer handles their data, applications and user access management, as well as the application's code, including the exposure of environment variables. It also says customers decide whether its platform and protections meet their application's needs. The customer sets the Deployment Protection scope, environment variable scope and firewall rules, and those settings are what the worksheet on this page checks.

It depends on the project's Deployment Protection scope. Vercel's environments page says new projects start with Standard Protection and Vercel Authentication unless the team set another default, so previews need a Vercel login while production domains stay public. Its generated URLs page says a deployment's unique URL is publicly accessible by default unless deployment protection makes it private. Shareable Links, Deployment Protection Exceptions and automation bypass secrets can also open a protected preview, so read the setting and the bypass lists for each project.

Vercel's platform-wide firewall with DDoS mitigation runs on every plan without configuration, and the Vercel WAF is available on every plan. The plan changes the limits: up to 3 custom rules per project on Hobby and up to 40 on Pro, no System Bypass Rules on Hobby, and the OWASP core ruleset only on the contract plan. The Bot Protection and AI Bots managed rulesets are on every plan but inactive by default.

Vercel's penetration testing guide says Pro customers and customers on its contract plan may test their own applications without advance notice for non-volumetric testing, and that testing is not permitted for Hobby users. Customers must not target Vercel's underlying hosting infrastructure and platform, and volumetric testing is limited to the contract plan with advance notice. Vercel's Fair Use Guidelines, which apply across plans, put penetration testing in the list headed "Never fair use" without naming a plan. Read both pages and your agreement on the day, and get Vercel's answer in writing where they leave doubt.

Runtime logs, which include Vercel Functions output from preview and production deployments, are kept for 1 hour on Hobby, 1 day on Pro and 3 days on the contract plan, or 30 days on Pro or the contract plan with Observability Plus, which shows up to 14 consecutive days at a time within that window. For anything longer, Vercel's shared responsibility model leaves retention to the customer through log drains, which are available on Pro and the contract plan.

Vercel's environments page answers no: previews use the variables scoped to Preview, with branch-specific values overriding them. It names one exception, the first deployment of a new project, which is always a production deployment built with Production variables. Two other cases put production values into a build people may treat as a preview: a staged production deployment, which uses production variables, and a preview promoted to production, which Vercel rebuilds with Production variables. A live credential added to Preview for all non-production branches reaches each of their builds unless a branch-specific value overrides it.

Scope the right test for your application

Tell us what you have built, who uses it and when it needs to launch. You will get a plain-English recommendation: a website audit, a web application test or a referral to a specialist.