QG301 Router bypass¶
| Type | Vulnerability |
| Severity | by what the action can do: Critical for a destructive or irreversible write, High for a write, Medium for a read, High when no extracted model classes it |
| Front end | a synthesised host and its router |
| Guardrail category (Vassilev 2026) | E. Action & Execution Guardrails |
| AI RMF functions (NIST AI 100-1) | GOVERN, MANAGE |
| OWASP | LLM03:2026 Excessive Agency, ASI02:2026 Tool Misuse and Exploitation |
| Gates by default | yes: a new finding under it fails the default quality gate |
Raised when¶
A tool reached outside the agent's whitelist, a query matching nothing resolved, or a tool run after the router resolved nothing.
Why it matters¶
The routing obligations are what keep a turn inside what an agent may do. One that fails dispatches a tool the agent was never given, or dispatches on no match.
How to fix it¶
Change the input that produces the guard the patch adds: the agent whitelist, or the intent library entry.
OWASP LLM03:2026, "Complete mediation": enforce the check in the rail, which decides from trusted state, rather than relying on the model to decide whether the action is allowed.
OWASP LLM03:2026, "Minimize tools": limit the tools this agent can reach to the ones its task needs.
What the finding carries¶
A location in the document the team edits, a fingerprint that survives unrelated edits, a reproduction (a transcript replayed against NeMo Guardrails for a policy, a dialogue case played through the selector for a router, or the gate's inputs reproduced by agreement), and, where the scan can produce one, a fix whose re-scan it has already run. See the catalogue for the other rules.