Summary
9Router through version 0.4.41 exposes sensitive user data via unauthenticated API endpoints that serve request logs and detailed AI conversation records.
Remote attackers can query these routes to enumerate paginated logs and retrieve full conversation histories, system prompts, tool calls, and user email addresses whenever the endpoints are reachable over the network.
Affected context
This vulnerability appears in deployments of 9Router where the request-logs and request-details API routes are accessible without authentication or authorization checks.
It is most relevant in environments where 9Router is exposed to untrusted networks (such as the public internet) and is used to proxy or record AI conversation data, including system prompts, user inputs, assistant outputs, tool invocation details, and associated user identifiers like email addresses.
Exploit preconditions
- Attack surface: HTTP(S) API endpoints for request-logs and request-details
- Required access: Network reachability to the 9Router instance; no valid account or credentials needed
- Trigger point: Sending crafted or enumerated requests to log listing and detail endpoints to paginate through stored records
- Why exploitation is possible: The affected routes lack authentication middleware and effective authorization checks (CWE-862), allowing untrusted callers to read data that should be access-controlled (CWE-359)
Why this happens
The issue stems from an authorization design flaw where log and detail endpoints are implemented without mandatory authentication and role-based checks.
From a CWE-359 perspective, sensitive information is stored and exposed via APIs that do not enforce access controls, causing unintended information disclosure.
At the same time, the absence or misplacement of authentication middleware maps to CWE-862, reflecting missing authorization controls on security-relevant functions.
Abstractly, the pattern is: a service that records detailed operational and user data exposes introspection or observability APIs, but treats them as benign diagnostic features rather than protected data interfaces.
Potential impact
- Attackers may access complete AI conversation histories, including user inputs and assistant responses.
- Attackers may retrieve internal system prompts, revealing configuration, business logic, or proprietary model orchestration details.
- Attackers could collect user email addresses and other identifiers, enabling phishing, account targeting, or correlation with other datasets.
- Attackers may infer tool usage patterns and operational metadata about the AI system, aiding further attacks or reconnaissance.
- Exposure of sensitive or regulated content processed by the AI system could lead to compliance violations and reputational damage.
General mitigation guidance
Effective mitigation typically focuses on treating log and observability endpoints as sensitive data surfaces and enforcing strict access control around them.
Common defensive patterns include:
- Requiring authentication and authorization for all endpoints that expose stored requests, logs, or conversation details, with least-privilege roles.
- Centralizing and consistently applying authentication middleware so that security-critical routes cannot bypass it.
- Minimizing the amount of sensitive data retained in logs by default and redacting user identifiers and message content where not strictly necessary.
- Segmenting administrative and diagnostic interfaces onto restricted network paths or management planes that are not exposed to general users or the internet.
- Applying data classification and privacy policies to API responses so that high-sensitivity data is never returned by default.
How teams detect issues like this
Detection often relies on a combination of design-time review and runtime monitoring:
- Static analysis or configuration review to ensure all routes that return logs, histories, or sensitive fields are annotated with authentication and authorization requirements.
- Code review practices that flag any controller or handler exposing user identifiers, message bodies, or system prompts where access control is missing or optional.
- Automated API contract checks that verify protected resources require appropriate security headers or tokens.
- Logging and monitoring of access to diagnostic and log endpoints for unusual patterns such as large-scale enumeration or unauthenticated queries.
- Policy enforcement in CI/CD that blocks deployment of new routes without explicit security metadata or guard middleware.

