runtime monitor

A runtime monitor is a software component that observes events while a program runs and compares them with properties specified in advance. It is the central tool of runtime verification and differs from static verification, which analyzes code before it executes, by targeting actual executions.

When it detects a violation, a monitor can be designed to do more than log or warn: it can return the system to a safe state or trigger a fallback procedure. In AI, runtime monitors are discussed as safeguards that keep the actions of AI agents, whose internals are hard to verify, within set bounds.

This entry is based on AIPOST articles and widely known facts. If something is wrong, please send us a correction request.