When the Agent Acts, Who's Accountable?

The Brief

For most of computing history, accountability was simple. A bug had an author. A bad query had a developer. There was always a name.

Agentic AI breaks that. An agent doesn’t answer a prompt — it plans, chooses tools, accesses systems, and executes multi-step actions on its own. When it’s wrong, the consequence isn’t a bad answer. It’s an unauthorized transaction, a modified record, a leaked dataset. And in a multi-agent system, where one agent orchestrates a dozen others, “who did this?” becomes a genuinely hard question.

This is not theoretical anymore. McKinsey found that 80% of organizations have already encountered risky agent behaviors — unauthorized data exposure, improper system access. Most of those organizations could not tell you which agent did it, or who owned that agent.

You can’t govern what you can’t name.

The Number

**80%** of organizations have already hit risky agent behaviors. Not “will.” Have. Meanwhile, Gartner projects that by 2030, half of all AI agent deployment failures will trace back to governance gaps rather than model failure.

*Source: McKinsey (2026 agent risk survey); Gartner 2026 Data & Analytics Predictions.*

The Move

Build an **agent inventory** this week. For every autonomous agent running in your organization, capture four fields:

1. **What it does** — the task and the workflows it spans

2. **What it can touch** — systems, data, and external tools it can reach

3. **What it can do without a human** — the actions it can take unsupervised

4. **Whose name is on it** — one accountable owner

If you can’t fill in column 4 for every row, you’ve just found your governance gap. Singapore published the first dedicated agentic-governance framework (the Model AI Governance Framework for Agentic AI) in January 2026 around exactly these ideas — autonomy assessment, human accountability, risk-bounding by design. The inventory is how you operationalize all three.

The Question

Brian Diamond is a fractional Chief AI Officer and founder of BrianOnAI, an AI governance platform, and Onaro, an AI spend intelligence platform. The CAIO Brief publishes every week for executives navigating AI leadership in real time.