When AI Agents Cross the Lines explores one of the most consequential shifts happening in cybersecurity: the transition from generative AI that provides information to agentic AI that can independently take action.

AI agents increasingly have many of the attributes security teams associate with highly privileged users: identities, credentials, access to sensitive information, connections to enterprise applications, and authority to execute transactions and change systems. But unlike human users, agents can operate autonomously, at machine speed, and across interconnected tools and environments.

That combination creates a fundamentally different security problem.

Recent real-world AI security incidents provide an early warning of what can happen when capable agents operate beyond their intended boundaries. They also expose an important gap: organizations have spent decades building security controls around applications, machines and human identities. Autonomous agents don't fit neatly into any of those categories.

Drawing on the work of the OWASP GenAI Security Project, including its agentic security research and guidance, this session translates emerging agentic threats into practical security considerations.

We will examine identity and privilege, tools and actions, goals and behavior, memory and context, data access, and detection and response. Finally, we'll connect these risks to the security programs organizations already have. Rather than creating another isolated AI security silo, organizations need to determine how identity, application security, data security, SecOps, and insider threat programs evolve to govern autonomous actors.

The takeaway: The question is no longer whether AI agents will have access to your enterprise. It's whether your security program can recognize when an authorized agent crosses the line—and stop it before autonomy turns into compromise.