← Ai4 2026 collection
Field Note 06Ai4 2026
Tracking AI Before It Escapes the Sandbox
Rajan Koo
Chief Technology Officer · DTEX Systems
AI Security · 9:11
Las Vegas · August 2026
Las Vegas · August 2026
How behavioral telemetry can reveal what an AI agent is doing before risky or malicious activity becomes a major incident.
Watch
The question we asked
“Can we tell what an AI agent is doing before it becomes a real security threat?”
The Field Note
As agents become more autonomous, conventional alerts often arrive after the important behavior has already occurred.
Behavioral monitoring can surface the weak signals, unusual sequences, and hidden activity that reveal both human and AI risk earlier in the process.
Three things to take away
- 01AI security depends on gathering the right behavioral telemetry.
- 02Warning signs often exist before a formal alert fires.
- 03Understanding agent behavior matters as much as inspecting its output.