In short
- Vercel's shared responsibility model makes you responsible for your application's code, the exposure of its environment variables and your app's own authentication.
- The April 2026 security incident is real and documented by Vercel. It started with the compromise of Context.ai, a third-party AI tool used by an employee, and exposed non-sensitive environment variables of a limited subset of customers. Vercel closed it on 16 October 2026.
- Store every credential as a Secret variable. Secrets are write-only after saving, which is the protection Vercel's bulletin points customers to.
- Many searches for a Vercel breach are about Next.js CVEs. Apps hosted on Vercel were not exposed to the 2025 middleware bypass, and Vercel's firewall filtered React2Shell, but only upgrading fixes a vulnerable version.
- The firewall does not know your data model. A function that returns another user's record passes through it like any normal request.
On this page
- Is Vercel secure? What it protects and what it leaves to you
- What Vercel gives you, by plan
- Has Vercel been breached? The April 2026 security incident
- What the incident changes for your project
- Why a search for a Vercel breach often lands on Next.js CVEs
- What stays yours on Vercel: four places apps leak
- How to check your Vercel project
- What Keybrum tests on an app hosted on Vercel, and what it does not
- Questions
- Sources
Is Vercel secure? What it protects and what it leaves to you
Vercel security has two halves, and Vercel says so itself. Its documentation publishes a shared responsibility model. Vercel manages the infrastructure: compute, storage and networking. The customer handles their data, their application and user access. The sentence that matters most for a small team is this one:
Customers are responsible for managing their application's code, including the exposure of environment variables.
| Area | Vercel's part | Your part |
|---|---|---|
| Network | A platform-wide firewall with automatic DDoS mitigation, free on all plans and with no configuration | Costs and safeguards for malicious traffic beyond that, spend limits, rate limits on your own endpoints |
| Transport | HTTPS on every deployment, with certificates generated automatically | Security headers such as a Content-Security-Policy |
| Compute | Isolation between customers' applications | What your functions do, and who may call them |
| Secrets | Environment variables encrypted at rest, a write-only Secret type, redaction in build logs | Choosing the type, never putting a secret behind a public prefix, rotating |
| Access to deployments | Deployment Protection: Vercel Authentication on all plans | Turning it on, reviewing exceptions, links and bypass secrets |
| Your app's users | Nothing: the docs say customers handle their app's authentication | Sign-in, sessions, and which user may read which record |
| Team accounts | Platform login, roles, SSO on higher plans | Multi-factor authentication for every member, least access |
So the honest answer to "is Vercel secure" is: the platform is run with care, holds a SOC 2 Type 2 attestation, and has still had an incident. And most leaks in apps hosted on Vercel have nothing to do with Vercel. They are in the app.
What Vercel gives you, by plan
- Firewall. DDoS mitigation is automatic for everyone. The Web Application Firewall (WAF, a filter in front of your app) is available on all plans, with up to 3 custom rules on Hobby and 40 on Pro. The OWASP core ruleset is listed for Enterprise only.
- Deployment Protection. Vercel Authentication, which requires a Vercel login with access to the project, is on all plans. Password Protection is $20 a month per protected project on Pro and not available on Hobby. Trusted IPs is Enterprise only.
- Variable types. Config values stay readable to members with access. Secret values are write-only after saving. When a Secret of 32 characters or more appears in a build log, Vercel replaces it with a redaction marker.
- Blocked vulnerable builds. Since 5 December 2025, new deployments that contain a Next.js version vulnerable to React2Shell fail to deploy unless you explicitly override the block.
- Security Review. A scanner that reads your code for vulnerabilities and reports findings in a dashboard.
These are real protections, and several of them cost nothing. None of them looks at what your routes return to a stranger. For how the same topics compare on a competitor, see Vercel vs Netlify.
Has Vercel been breached? The April 2026 security incident
Yes. Vercel published a security bulletin titled "Vercel April 2026 security incident" and updated it from 19 April 2026 until it closed the incident on 16 October 2026. Here is what the bulletin itself says. We leave out what only press articles reported.
- 1
The entry point was a third-party tool
The bulletin states that the incident originated with a compromise of Context.ai, a third-party AI tool used by a Vercel employee. The attacker used it to take over the employee's Google Workspace account, then their Vercel account.
- 2
The attacker read non-sensitive environment variables
From there the attacker moved into a Vercel environment and, in the bulletin's words, maneuvered through systems to enumerate and decrypt non-sensitive environment variables. Those are the variables whose stored value can be decrypted back to plain text.
- 3
A limited subset of customers was affected
Vercel first identified a limited subset of customers, contacted them and recommended an immediate rotation of credentials. It later identified a small number of additional accounts compromised as part of the incident, and notified them. The bulletin gives no count.
- 4
Vercel brought in outside help
The bulletin says Vercel worked with Google Mandiant, other cybersecurity firms, industry peers and law enforcement, and published an indicator of compromise for the wider community: the identifier of the Google Workspace OAuth app involved.
- 5
The npm packages were checked
Vercel publishes Next.js and other widely used packages. On 20 April it stated that no npm packages published by Vercel had been compromised.
- 6
The incident was closed in October
The entry of 16 October 2026 reads: all remediation actions related to the security incident are complete, and the incident is closed.
One more point from the bulletin is easy to misread. On 23 April Vercel reported a small number of customer accounts with signs of compromise that, it said, did not originate on Vercel systems and did not appear to be a continuation of the April incident.
What the incident changes for your project
The lesson is narrower than "leave Vercel". Any host can have an employee account taken over. What decided the damage for a customer was the type of each variable. A value stored as readable could be decrypted. Vercel's recommendations, from the bulletin:
- Enable multi-factor authentication on every team member's account, with an authenticator app or a passkey
- Rotate the credentials that were stored as non-sensitive variables, as a priority
- Store secrets as sensitive values (the Secret type today) so they cannot be read back
- Review the activity log and recent deployments for anything you do not recognise
- Set Deployment Protection to Standard at a minimum, and rotate Deployment Protection tokens if you use them
After the incident Vercel shipped stronger defaults for environment variables, a team-wide view of them and an easier activity log. The current docs describe the two types, Config and Secret, and an optional team policy that requires a separate Secret value for Production. There is a wider point for any company: the entry was a third-party AI tool connected to one employee's work account. Review which tools hold access to yours.
Why a search for a Vercel breach often lands on Next.js CVEs
Vercel maintains Next.js, so its name is attached to every Next.js vulnerability. Those are flaws in a framework you install, whichever host runs it. Two of them explain most of the search results.
| Vulnerability | What it allowed | Apps hosted on Vercel |
|---|---|---|
| CVE-2025-29927, CVSS 9.1, March 2025 | Skipping Next.js middleware with one internal header, so pages protected only by middleware opened with no session | Not exposed: Vercel's routing runs in a separate system. Self-hosted apps were |
| CVE-2025-55182, React2Shell, CVSS 10.0, December 2025 | Running code on the server of an unpatched App Router app with one request | Filtered by firewall rules Vercel deployed for all users, and vulnerable versions blocked from new deployments. Vercel states that the rules cannot cover every variant and that upgrading is the only complete fix |
The pattern repeats with newer advisories: the host can filter, and the fix is still a version bump in your package.json. The Next.js security guide lists the current floor and every CVE we cover.
What stays yours on Vercel: four places apps leak
- Public-prefixed variables. Vercel's docs note that Next.js inlines
NEXT_PUBLIC_variables into the client bundle at build time. A secret behind that prefix is published to every visitor, however it is stored. See API key exposed in the frontend. - Functions and Server Actions. Each one is an endpoint on the internet. The firewall lets a well-formed request through, and only your code can check the session and the owner of the record. That gap is an IDOR.
- Previews. Old previews keep running old code and old variables. Protection exceptions, shareable links and automation bypass secrets are doors you opened and may have forgotten.
- Headers. Vercel's production checklist asks you to implement a Content-Security-Policy and proper security headers. They are not added for you.
How to check your Vercel project
- 1
Read the type of every variable
In the project, open Environment Variables. Each row is labelled Config or Secret. Every password, API key and token should be a Secret. For a credential stored as Config, rotate it at its provider and save the new value as a Secret, since the old one was readable.
- 2
List what the build publishes
Run the first command below in your repository and read each name. Then search the built files for secret key formats.
- 3
Open Deployment Protection
In Settings, then Deployment Protection, confirm Vercel Authentication with Standard Protection or stricter. Review the exceptions, the shareable links and the automation bypass secret.
- 4
Test a preview while logged out
Request a preview URL from a terminal or a private window. A protected preview answers with a login page.
- 5
Check the installed Next.js version
Run
npm ls next react react-domand compare with the fixed versions in the Next.js security guide. - 6
Review the team
Remove members who left, check roles, and confirm multi-factor authentication for everyone. Then read the activity log for the past weeks.
# Variables the build copies into the browser bundle
grep -rhoE "NEXT_PUBLIC_[A-Z0-9_]+" app src pages components 2>/dev/null | sort -u
# Secret formats in the built client files
npx next build
grep -rlE "sk_live_|sk-proj-|sb_secret_|service_role|BEGIN (RSA |EC )?PRIVATE KEY" .next/static
# Is a preview public? 200 means anyone can open it.
curl -s -o /dev/null -w "%{http_code}\n" "https://YOUR-PREVIEW.vercel.app"
# Installed versions, from the lockfile
npm ls next react react-domWhat Keybrum tests on an app hosted on Vercel, and what it does not
Keybrum is the automated pentest we build for apps made with AI. It attacks the deployed app from outside, once its owner has proved ownership. Vercel's Security Review reads your code. Keybrum tests what the live address answers, which is where a missing check shows.
- Leaked keys: secrets of many providers in the JavaScript and HTML the app serves, each checked with a read-only call.
- Routes and actions: every API route your JavaScript calls, with no session and then across two accounts, to find records one user can read from another.
- Next.js and React CVEs: proved by behaviour, including the middleware bypass and React2Shell. Keybrum never runs code-execution or denial-of-service payloads: it proves the flaw is reachable.
- Served files and headers:
.envfiles, source maps, and the baseline of security headers. - The backend behind the app: Supabase or Firebase rules, tested with the public key.
On paid plans, connecting Vercel lets it attack each deploy, previews included, before it ships, and the Team plan (€99 a month excluding VAT, 25 apps) finds every app deployed on your Vercel, Netlify or Supabase accounts. A one-time report on one app is €49, and Maker is €29 a month for 5 apps.
It does not audit your Vercel account: member roles, variable types, the activity log and protection settings are yours to read. It would not have detected the April 2026 incident, which happened inside Vercel's systems. And a clean run says which attacks failed on which date, nothing more.
Questions
Was Vercel hacked in 2026?
Vercel disclosed a security incident in April 2026. According to its bulletin, an attacker compromised a third-party AI tool used by an employee, took over that employee's accounts and decrypted non-sensitive environment variables of a limited subset of customers. Vercel declared the incident closed on 16 October 2026.
Were my Vercel environment variables exposed?
Vercel says it contacted the affected customers directly. The bulletin describes the exposed values as non-sensitive environment variables, the type that can be decrypted to plain text. If you stored credentials in readable variables at the time and are unsure, rotate them and store the new values as Secrets.
Is Vercel safe to host a production app?
For most teams, yes. The platform provides HTTPS, DDoS mitigation, a firewall and preview protection, and holds a SOC 2 Type 2 attestation. What it cannot do is check your application's logic: who may call a function and whose data it returns.
Does the Vercel firewall protect my app from attacks?
It filters network-level attacks automatically and lets you add rules, and Vercel has shipped rules against specific Next.js and React flaws. It does not replace upgrades, and it cannot tell that a valid request asks for someone else's record.
Was Vercel affected by the Next.js middleware bypass, CVE-2025-29927?
Apps hosted on Vercel were not exposed, because Vercel's routing runs in a separate system. Self-hosted Next.js apps on vulnerable versions were. If you moved an app off Vercel since, check its Next.js version.
What should I do after a host reports an incident?
Rotate every credential the host stored in a readable form, at the provider that issued it. Turn on multi-factor authentication, review the activity log and recent deployments, and move secrets to the write-only type. Deleting the project does not revoke a key.
Sources
- 1.Vercel security overview, Vercel Docs
- 2.Shared Responsibility Model, Vercel Docs
- 3.Vercel April 2026 security incident, Vercel
- 4.Config and Secret environment variables, Vercel Docs
- 5.Deployment Protection on Vercel, Vercel Docs
- 6.Vercel WAF, Vercel Docs
- 7.Production checklist for launch, Vercel Docs
- 8.New deployments of vulnerable Next.js applications are now blocked by default, Vercel
- 9.React2Shell security bulletin, Vercel
- 10.Postmortem on Next.js middleware bypass, Vercel