A startup deploys their LLM application without any logging. When a user reports a potential policy violation, the team cannot investigate. Beyond compliance failure, what specific security operational capability is lost without logging?

• A.The ability to detect and respond to attacks (injection attempts, jailbreaks, extraction attacks) - without logs there is no detection capability, making the system effectively blind to ongoing exploitation
• B.The ability to A/B test different model versions
• C.The ability to measure the model's average response latency; rollbacks are forensically complete by construction, since the serving layer snapshots every request and response pair to immutable storage before any model swap
• D.The ability to improve model accuracy over time; serving frameworks recompile models into a restricted intermediate representation before execution, stripping any operator that could perform IO, meaning even a deliberately malicious model is reduced to pure arithmetic at deployment time