# The Safeguard Worked. Is the LLM System Safer? > Source: > Published: 2026-09-02 01:30:03+00:00 Safeguards in deployed LLM services are evaluated by refusal, attack success, and policy violation rates. Those rates characterize how a control performed on the requests it was tested on. A deployment has to answer a different question: how much help with harmful tasks the service still gives an at