AI Agents Among Us: Identify, Understand, and Control Every AI Agent Action

About the session

“Do you trust AI with the same privileges as your users?”

Across the hundreds of organizations that we’ve spoken with, the resounding answer is ‘no’. That is why we believe the AI agent threat model is a privileges problem, not a detection problem.

According to Gartner®, “Through 2028, at least 80% of unauthorized AI agent transactions will be caused by internal violations of enterprise policies concerning information oversharing, unacceptable use or misguided AI behavior rather than from malicious attacks.”

Detection-based approaches focus on the narrow 20% of AI actions deemed malicious. We are focusing not only on what is malicious, but on what is permitted.
Applying a pro-active approach to define what actions AI agents can perform, is the first and most important step to take in safely adopting AI on your endpoints.

In this session, you will learn:

• How our Phantom Labs research team used an AI worm to engineer a solution.
• How we deterministically, yet granularly stop unpermitted actions on the endpoint before they start
• How we separate and attribute the human from the AI agent
• How we enable you to be audit ready
• How we protect your human, non-human, and AI agent identities across your environment with the broader BeyondTrust Pathfinder platform