Continuous Monitoring

Continuous Monitoring refers to a framework that keeps recording and watching what an AI system does while it operates. It's similar to a CCTV camera mounted on a store ceiling that constantly films who came in and what they touched. Rather than investigating only the moment something goes wrong, the system keeps recording all along, so the footage can be reviewed later when needed.

This kind of logging matters even more for AI that can run external programs or connect to the internet on its own. It's not enough to just keep a record of what was asked and what answer was given. To trace the cause of an incident, you also need to know which external program was called, what values were used to run it, and where it ended up connecting as a result.

When regulators require continuous monitoring, they're essentially telling operators to prepare in advance the evidence needed to reconstruct what happened after an incident occurs. It marks a shift from managing only outcomes to managing the entire process.

In articles, it shows up in phrasing like "high-risk systems require continuous monitoring." This means a surveillance mechanism that must be running at all times, from before any incident occurs, rather than a corrective measure taken after the fact. It's easy to confuse with incident investigation, but the key point of continuous monitoring is that it operates constantly, regardless of whether an incident has occurred.