Vision Paper: Governing AI Agents

Why this matters

From tools that suggest, to agents that act

AI is moving from systems that produce an output for a human to review, to agents that plan, call tools, and complete multi-step tasks with growing autonomy. That shift raises a new operational question for every risk function: how do you keep these systems safe, reliable, and aligned with your institution's objectives, at the pace the business wants to deploy them?

The underlying risks aren't new. What's genuinely new is the scale, and that's what breaks a largely manual governance process. The paper explains why, and what to do about it.