/

CVE-2026-73264

Severity:

HIGH RISK

(7.6)

SSRF (CWE-918) in Prowler Lighthouse OpenAI Provider

CVE-2026-73264

Awaiting Analysis

Vulnerability type

Category:

Server-Side Request Forgery

CWE:

CWE-918

Impact:

Server-Side Request Forgery and API Key Disclosure

Severity:

HIGH RISK

(7.6)

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

Prowler contains a server-side request forgery vulnerability in the Lighthouse OpenAI-compatible provider configuration. An authenticated user with provider configuration access may cause the service to send requests, including the configured API key, to attacker-controlled or internal endpoints when listing models.

Affected context

The issue occurs in Prowler deployments where Lighthouse providers can be created or updated through the API and the openai_compatible provider accepts a user-controlled base_url. Risk is greater when the Prowler service can reach internal network services or when the configured provider credential is included in outbound authorization headers.

Exploit preconditions

  • Attack surface: Lighthouse provider configuration API endpoints
  • Required access: Authenticated access with permission to configure Lighthouse providers
  • Trigger point: Model listing through the configured OpenAI-compatible provider
  • Why exploitation is possible: The supplied base_url is not sufficiently validated before the backend sends an outbound request and forwards the provider API key

Why this happens

The vulnerability results from accepting a provider endpoint from an authenticated user without applying adequate URL validation, destination restrictions, or trust-boundary enforcement. The backend then uses that endpoint for server-side network communication and attaches a sensitive authorization header, mapping to CWE-918: Server-Side Request Forgery. This insecure pattern allows application-controlled requests to reach destinations that should not be selectable by configuration input.

Potential impact

  • Requests to attacker-controlled systems may expose the configured API key
  • Internal services may receive requests originating from the Prowler server
  • Cloud metadata or administrative endpoints could be probed where network access permits
  • Internal service information may be disclosed through response handling or operational logs
  • Downstream systems may be affected if requests are accepted as trusted

General mitigation guidance

Mitigation generally focuses on reducing trust in user-supplied provider URLs and restricting outbound network access.

Common defensive patterns include:

  • Allowlisting approved provider schemes, hosts, ports, and deployment-specific destinations
  • Resolving and validating destinations before connection, including protections against DNS rebinding and private-network targets
  • Avoiding automatic forwarding of API keys to destinations that are not explicitly trusted
  • Separating connection testing from privileged provider operations and minimizing returned response data
  • Applying egress filtering, network segmentation, and least-privilege service connectivity
  • Keeping provider credentials narrowly scoped and preventing their exposure in logs or error messages

How teams detect issues like this

Organizations can look for SSRF risks through:

  • Static analysis rules identifying user-controlled URLs passed to HTTP clients
  • Code review checks for missing scheme, host, port, redirect, and private-address validation
  • Review of code paths that attach credentials to requests based on configurable endpoints
  • API policy enforcement restricting provider URLs to approved destinations
  • Network telemetry monitoring unexpected connections from Prowler to internal, metadata, or unusual external hosts
  • Audit logging of provider configuration changes and outbound connection targets

Automated security review tools may help identify URL-to-request flows and credential forwarding patterns.

Sources

SourceLink
github.comgithub.com/prowler-cloud/prowler/commit/1c4d8e3e756ee04367353101af797004d1c0258f
github.comgithub.com/prowler-cloud/prowler/commit/3f2e5929d9ff3c1a5d2fae49f8edd419ee99a8e2
github.comgithub.com/prowler-cloud/prowler/pull/11928
github.comgithub.com/prowler-cloud/prowler/pull/11940
github.comgithub.com/prowler-cloud/prowler/releases/tag/5.33.1
github.comgithub.com/prowler-cloud/prowler/releases/tag/5.34.0
github.comgithub.com/prowler-cloud/prowler/security/advisories/GHSA-fhj4-q47f-w7mv

Frequently asked

questions

It can be exploitable through the Prowler API when an attacker has authenticated access allowing Lighthouse provider configuration changes. It is not necessarily reachable by unauthenticated internet users.

The attacker needs permission to create or modify Lighthouse provider configurations, including the OpenAI-compatible provider endpoint.

Treat provider URLs as untrusted input, restrict them to an explicit allowlist, block private and metadata destinations, and avoid sending credentials to unapproved endpoints.

Combine strict URL validation with outbound network controls, least-privilege credentials, configuration-change auditing, and code review rules for user-controlled URLs used by HTTP clients.

Find out what's already
exploitable in your codebase.