Summary
A missing authorization check in Dradis Community Edition allows any authenticated non-admin user to configure an AI provider that points to an arbitrary HTTP or HTTPS address and assign it to the built-in Roslin agent. When an AI interaction is triggered, the server may send a request to the attacker-selected destination, and some response bodies may be returned to the attacker through ActionCable or Turbo Stream errors.
Affected context
The issue occurs in Dradis Community Edition deployments where the ProvidersController and AgentsController conditionally apply administrative authorization based on the presence of the Dradis::Pro constant. Because that constant is not defined in the Community Edition context, the intended authorization gate may not run. The risk is greatest when authenticated users can create or modify AI providers, reassign agents, and trigger AI interactions from a server that can reach internal or restricted network destinations.
Exploit preconditions
- Attack surface: Authenticated web application functionality for AI providers and agents
- Required access: Any authenticated account with access to the affected provider or agent workflows; administrative privileges may not be required
- Trigger point: An AI interaction that causes the server to contact the configured provider URL
- Target conditions: The server must be able to resolve and connect to the selected HTTP or HTTPS destination
- Why exploitation is possible: Authorization is conditionally skipped, allowing an untrusted user-controlled URL to reach a server-side request mechanism
Why this happens
The underlying bug is an authorization control-flow error. Administrative protection is gated on a feature constant that is absent in the Community Edition, causing the intended before_action to be omitted rather than enforced. This allows untrusted users to influence a server-side network request destination, which maps to CWE-918: Server-Side Request Forgery. Reflecting certain upstream response bodies through application error messages can additionally expose information obtained by the server.
Potential impact
- An attacker may make the server send requests to internal, private, or link-local services
- Internal service metadata or response content could be disclosed when responses are reflected
- Cloud instance or platform credentials could be exposed if reachable through metadata endpoints
- The server could be used to probe internal network availability and service behavior
- Requests may reach administrative interfaces that could enable further compromise, depending on network controls
General mitigation guidance
Mitigation generally focuses on enforcing authorization independently of optional product constants and treating provider destinations as untrusted input. Common defensive patterns include applying explicit permission checks to provider creation, provider modification, and agent reassignment; using an allowlist of approved provider hosts or destinations; blocking loopback, private, link-local, and other restricted address ranges after DNS resolution; validating redirect destinations; and routing outbound requests through a controlled egress proxy. Response handling should avoid reflecting arbitrary upstream bodies to users, while network policies, least-privilege service accounts, and restricted egress can limit impact when application controls fail.
How teams detect issues like this
Detection can include static analysis and code review for authorization callbacks guarded by optional constants, feature flags, or environment-dependent conditions. Reviewers can look for user-controlled URLs passed to HTTP clients, missing destination allowlists, unrestricted redirects, and error handlers that return upstream response bodies. Runtime monitoring can alert on requests from the application to private, loopback, link-local, or unexpected external ranges, as well as changes to AI provider and agent assignments. Policy enforcement may require centralized authorization checks, outbound proxying, and automated tests covering non-admin access to administrative workflows.

