Large language model (LLM) routers are increasingly used to manage trade-offs between quality, latency, and cost by dynamically choosing which model to invoke and rewriting requests. But this convenience introduces a new attack surface: if a router is compromised, it can make ineligible selections that undermine the intended control.

The arXiv paper "VeriRouter: Towards Verifiable LLM Routing" highlights this risk and proposes a framework called VeriRouter. The abstract warns that a compromised router can cause ineligible selection, but it does not yet describe the underlying mechanism or proof approach in the available text.

As a single-source digest, this summary reflects only the abstract's stated motivation and direction. The full paper likely details how VeriRouter achieves verifiability, but that information is not present in the source provided.