Webinar Description
Key Takeaways
- Focuses on auditing, monitoring and risk-scoring autonomous AI agents operating in production environments
- Demonstrates Teleport’s Beams trusted runtime for agent session visibility and behavioural classification
- Addresses risk mapping against the MITRE ATT&CK framework for AI-driven activity
- Explores LLM proxy-level auditing across platforms including Claude Code, OpenCode, Pi and Codex
- Designed for security engineers, platform teams, IT architects and CISOs managing agentic AI deployments
Introduction
As organisations accelerate their deployment of autonomous AI agents, a critical operational gap has emerged: how do security teams maintain visibility into what these agents actually do? Unlike traditional software systems with predictable execution paths, AI agents make decisions dynamically, interact with infrastructure in ways that may not be fully anticipated, and can take actions with significant consequences before human operators have an opportunity to intervene.
This webinar from Teleport addresses these challenges directly, presenting new capabilities within its Identity Security platform designed specifically for organisations grappling with agentic AI governance. The session is particularly timely given the rapid proliferation of coding assistants, autonomous DevOps agents and AI-driven infrastructure management tools across enterprise environments.
About This Event
Titled “Agents as Toddlers: How We Get Audit and Visibility Into Risky Agent Action,” this live virtual webinar is led by a Teleport product director and combines technical demonstration with practical guidance. The session walks through Teleport’s Beams trusted runtime, showing how security and platform teams can implement comprehensive audit trails, behavioural monitoring and automated risk assessment for AI agents operating within their infrastructure.
Auditing Agent Sessions and Behaviour
A central theme of the webinar is the challenge of understanding agent reasoning and decision-making after the fact. Traditional logging approaches capture system calls and API requests but fail to explain why an agent took a particular action or whether that action aligned with its intended purpose. Teleport’s Beams Session Summaries aim to bridge this gap by providing structured records of agent activity that security teams can review and analyse.
The session also covers Agentic Classifiers, which monitor agent behaviour in real time and categorise actions according to predefined risk profiles. This classification layer enables organisations to distinguish between routine operations and potentially problematic behaviour without requiring manual review of every agent interaction.
Risk Scoring and Framework Alignment
For organisations operating in regulated industries or managing critical infrastructure, the ability to map agent activity against established security frameworks is increasingly important. The webinar demonstrates how Teleport’s risk scoring capabilities align agent actions with the MITRE ATT&CK framework, providing a common vocabulary for assessing potential threats and communicating risk to stakeholders.
This approach allows security teams to triage agent-related incidents more efficiently, focusing attention on high-risk behaviours while allowing lower-risk activity to proceed with appropriate logging. As the volume of agent-driven operations scales, automated risk assessment becomes essential for maintaining security posture without creating operational bottlenecks.
Zero Trust in an Agentic World
The webinar positions zero trust architecture as a necessary foundation for agentic AI security while acknowledging its limitations. When machines become autonomous actors capable of making consequential decisions, identity verification and access control alone are insufficient. Organisations must also understand what agents are doing with their access and whether those actions fall within acceptable boundaries.
LLM proxy-level auditing across platforms such as Claude Code, OpenCode, Pi and Codex extends this visibility to the model layer, capturing the prompts, responses and tool invocations that drive agent behaviour. This granular audit trail supports both security investigations and compliance requirements.
Who Should Attend
This webinar is designed for security engineers, DevOps practitioners and platform engineers responsible for deploying and managing AI agents in production. IT security leaders, architects and CISOs evaluating governance frameworks for agentic AI will find the session relevant to their strategic planning. Organisations in regulated sectors or those managing critical infrastructure face particular urgency in establishing robust agent oversight capabilities.