/

CVE-2026-85699

Severity:

HIGH RISK

(7.5)

Server-Side Request Forgery (CWE-918) in jina-ai Reader

CVE-2026-85699

Deferred

Vulnerability type

Category:

Server-Side Request Forgery

CWE:

CWE-918

Impact:

Unauthorized access to internal services and sensitive network responses

Severity:

HIGH RISK

(7.5)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

A server-side request forgery vulnerability in the jina-ai reader allows a crafted public URL to redirect the server toward internal network services or cloud metadata endpoints. When redirect destinations are not validated again, an attacker may cause the server to fetch and return responses from otherwise inaccessible resources.

Affected context

The issue occurs in jina-ai reader deployments that retrieve user-supplied or attacker-influenced URLs and follow HTTP redirects. It becomes risky when validation is applied only to the initial URL, while subsequent redirect destinations can resolve to private network addresses, loopback services, or cloud metadata endpoints.

Exploit preconditions

  • Attack surface: Server-side URL fetching performed by the jina-ai reader
  • Required access: Ability to submit or influence a URL processed by the reader
  • Trigger point: Following an HTTP redirect to a new destination
  • Why exploitation is possible: The initial URL is validated, but redirect targets are not subjected to equivalent network-location and destination checks

Why this happens

The vulnerability results from incomplete URL validation across a multi-request redirect flow. A server may correctly assess the first URL while implicitly trusting later redirect locations, allowing an external resource to steer the request toward internal addresses.

This maps to CWE-918: Server-Side Request Forgery, where attacker-controlled input causes a server to initiate requests to unintended destinations.

Potential impact

  • An attacker may access responses from internal services
  • Cloud instance metadata or temporary credentials could be exposed
  • Internal network topology and service details could be disclosed
  • Sensitive data returned by reachable services could be relayed to the attacker
  • Internal endpoints could be probed or abused through the vulnerable server

General mitigation guidance

Mitigation generally focuses on treating every redirect destination as untrusted and enforcing destination controls throughout the complete request chain.

Common defensive patterns include:

  • Re-applying URL, hostname, scheme, and IP-range validation after every redirect
  • Blocking loopback, link-local, private, reserved, and cloud metadata address ranges unless explicitly required
  • Restricting outbound requests to an allowlist of approved schemes, hosts, and ports
  • Resolving hostnames carefully and checking the resulting addresses to reduce DNS-based bypasses
  • Limiting response exposure, redirects, timeouts, and outbound network permissions
  • Running the reader with minimal network privileges and isolating it from sensitive infrastructure

How teams detect issues like this

High-level detection can include:

  • Static analysis for URL fetchers that validate only the initial request
  • Code review checks for redirect handling without destination revalidation
  • Tests covering redirects to loopback, private, link-local, and metadata address ranges
  • Egress policy monitoring for unexpected connections from reader services
  • Logging of redirect chains, resolved addresses, schemes, and destination ports
  • Policy enforcement requiring outbound allowlists and SSRF-focused security tests

Automated code review and application security testing tools may help identify incomplete redirect validation patterns.

Sources

SourceLink
github.comgithub.com/jina-ai/reader
github.comgithub.com/jina-ai/reader/blob/1574bfd380d249c86c82db4dace0d9c8fe17e2b1/src/services/curl.ts
github.comgithub.com/jina-ai/reader/blob/1574bfd380d249c86c82db4dace0d9c8fe17e2b1/src/services/puppeteer.ts
github.comgithub.com/jina-ai/reader/issues/1252
vulncheck.comvulncheck.com/advisories/jina-ai-reader-server-side-request-forgery-via-redirect-validation-bypass

Frequently asked

questions

It may be remotely exploitable when an attacker can submit a URL that the jina-ai reader retrieves and the service can reach internal or cloud-provider endpoints.

No. The vulnerable server performs the request on the attacker's behalf, although exploitation depends on the server's outbound network access and the reader's URL-fetching interface.

Validate every redirect destination using strict scheme, hostname, port, and resolved-address rules, preferably with an allowlist of approved destinations.

Treat all URL components and redirect targets as untrusted, restrict outbound network access, test private-address and metadata protections, and monitor redirect chains and egress traffic.

Find out what's already
exploitable in your codebase.