Agentic AI Governance: Extending Frameworks Beyond GenAI

Agentic AI systems introduce a new governance risk that current GenAI programs often don't address. These systems can take approved actions across multiple systems in unsafe sequences, leading to unintended outcomes. Here's the thing: Many financial institutions have strong AI governance, but agentic AI breaks assumptions in existing frameworks. What's interesting is that the risk emerges from a chain of individually acceptable steps, not necessarily flawed outputs. This means treating agentic AI as just a more capable language model is a mistake. The bottom line is that governance needs to extend beyond just model outputs. This involves classifying agents by what they can do, redesigning human oversight around actions, and validating behavior across full workflow scenarios. This is crucial for managing the unique risks of these advanced AI systems.