Agentic AI & autonomous workflows: New governance risks and guardrails
July 22 @ 2:00 pm – 3:00 pm EDT
Description:
Agentic AI presents unique governance challenges. This webinar provides a practical model for where, how, and when to place guardrails across the agent lifecycle. We will dissect how autonomous workflows create new exposures — cascading errors across multi-step chains, ambiguous accountability when an agent acts, excessive or unintended permissions, prompt and tool-poisoning attacks, and the difficulty of reconstructing what an agent did and why.
Drawing on emerging practice, we will map these risks to concrete control placements: pre-action authorization gates, scoped permissions and least-privilege tool access, human-in-the-loop checkpoints calibrated to action severity, behavioral monitoring and circuit breakers, and audit-grade logging of agent reasoning and decisions.
Rather than treating agentic AI as either a novelty to ban or a black box to accept, this session equips risk, compliance, and internal audit professionals with a structured way to evaluate autonomous workflows, identify the control points that matter most, and design proportionate guardrails that preserve the value of automation without surrendering oversight. Attendees will leave with a practical framework they can apply to agentic deployments already emerging in their organizations.
Learning objectives:
- Identify the principal governance risks unique to autonomous workflows, including cascading and compounding errors, permission and privilege escalation, accountability gaps, adversarial manipulation of tools and instructions, and reduced explainability.
- Map control points across the agent lifecycle, determining where guardrails belong — at planning, authorization, tool invocation, action execution, and post-action review — and why placement matters.
- Discuss technical and procedural guardrails, including least-privilege tool scoping, pre-action approval gates, behavioral monitoring, kill-switches and circuit breakers, and audit-grade logging of agent reasoning and decisions.
- Evaluate accountability and ownership structures, applying the three-lines model to clarify who is responsible when an autonomous agent causes harm or error.
- Assess an agentic deployment in their own organization using a structured evaluation approach to surface gaps and prioritize guardrail investments.
CPE credit(s): 1 CPE upon live viewing and participation. CPEs not offered on-demand.
Field of study: Information Technology
Instructional delivery method: Group Internet Based
Level: Basic
Prerequisite: None
Advanced preparation: None
*Click here for more CPE information.
