Understanding Human, Agent, and Total Identities
An identity is anything that acts on an AI application. Some are people. Some are AI agents working on a person's behalf. The summary cards count each kind separately.
Splitting the two matters because they carry different risk. A person sending sensitive data to an AI tool made a decision. An agent doing the same acted on instructions someone gave it earlier, possibly without realising what it would send.
Table 1. What each card counts
| Card | What it counts |
|---|---|
| Total Interactions | Every exchange with the application in the reporting period. This counts activity, not identities, so it is normally far higher than the other three. |
| Human Identities | People who used the application directly. |
| Agent Identities | AI agents that acted on the application. |
| Total Identities | How many distinct identities were involved altogether. |
Reading them together
Agent Identities at zero means every interaction came from a person typing. That is the usual picture for a chat assistant used directly.
A high agent count against a low human count means most of the activity is automated. Sensitive data reaching the application in that case is coming from whatever the agents were told to do, so the place to look is the agent's configuration rather than any individual user's behaviour.