Summary
A server-side request forgery vulnerability in the Jina AI Reader allows unauthenticated attackers to make the service issue requests to attacker-selected destinations. The risk arises when publicly resolvable hostnames point to private addresses and the reader's private-address guard is disabled outside Google Cloud deployments.
Affected context
The vulnerability can affect Jina AI Reader deployments operating outside Google Cloud environments when their request filtering does not block hostnames that resolve to private or otherwise restricted IP addresses. It becomes relevant when the reader accepts attacker-controlled URLs or host references without requiring authentication.
Exploit preconditions
- Attack surface: A network-facing Jina AI Reader request or URL-fetching
interface
- Required access: No authentication, where the affected interface is
publicly reachable
- Trigger point: Processing an attacker-supplied hostname or URL - Why exploitation is possible: The private-address guard is disabled in
applicable non-Google Cloud deployments, allowing DNS resolution to lead
requests toward private addresses and internal resources
Why this happens
The underlying bug is an incomplete server-side request validation policy. The application conditionally disables protection against private-address destinations based on the deployment environment, while still accepting attacker-controlled hostnames. Because a public hostname can resolve to a private IP address, this behavior permits the server to act as a proxy to internal services. This maps to CWE-918: Server-Side Request Forgery.
Potential impact
- An attacker may retrieve cloud instance metadata
- An attacker could read content from internal services
- Internal network topology or service information may be disclosed
- Credentials or tokens exposed through metadata services could be misused
- Internal endpoints may be probed or abused through the vulnerable server
General mitigation guidance
Mitigation generally focuses on enforcing destination validation consistently across all deployment environments. Secure-by-default patterns include blocking loopback, link-local, private, multicast, and other restricted address ranges after DNS resolution; validating every redirect and resolved address; and applying strict URL scheme, port, and protocol policies. Reducing trust in user-supplied hostnames, using controlled outbound proxies or allowlists, isolating the service network, and assigning minimal runtime privileges can limit impact when filtering is bypassed. Cloud metadata services should also require strong request protections where supported.
How teams detect issues like this
Detection approaches include static analysis and code review for attacker-controlled URL fetches, environment-dependent disabling of SSRF protections, DNS resolution followed by unchecked outbound connections, and redirect handling that skips destination validation. Runtime monitoring can flag requests to loopback, private, link-local, metadata, and unexpected internal address ranges. Egress policies, proxy allowlists, network sensors, and automated security review rules can help enforce these controls.

