Mariano O'Kon
All field notes

Valid access. Dangerous behavior.

. 2-minute read.

Graphic: Valid access. Dangerous behavior.

An employee signs in with valid credentials.

An AI agent uses approved tools.

Both pass the access checks.

Then they start downloading unusual volumes of sensitive data or contacting unfamiliar external systems.

Would your security team recognize the change?

Behavioral analysis helps answer a simple question: Is this activity consistent with how this person, device, or AI agent normally works?

Darktrace’s November 18 article on Microsoft Perception highlights this distinction.

Identity and permissions establish who has access.

Behavioral analysis adds context about how someone uses access.

For AI agents, this matters even more.

An agent following instructions from a manipulated document might use legitimate permissions to perform harmful actions.

My takeawayEvaluate security by how quickly your team recognizes and contains harmful behavior, including activity using legitimate access.
  • Connect signals across your environment.
  • Investigate unusual activity in business context. Unusual does not automatically mean malicious.
  • Define which responses run automatically and which require human approval.
  • Test scenarios involving compromised employees and manipulated AI agents.

Ask your security leader: If an approved AI agent started misusing legitimate access, how would we detect and stop the activity?

Sources

Originally posted on LinkedIn