When AI Agents Make Costty Mistakes, Who Actually Takes the Blame?

According to data highlighted by the Olhar Digital columnist Alessandra Montini, 40% of large organizations with revenues over $1 billion are already scaling AI agents. Yet, research by Deloitte reveals that a mere 21% of companies possess a mature governance model to handle this new wave of autonomous technology.

Why autonomous actions change corporate risk

During the early wave of generative AI, the primary fear was factual inaccuracy, such as a chatbot producing a wrong statement. Autonomous agents, however, interact directly with internal systems, execute operational tasks, and trigger sequences of actions without constant human oversight.

A misinterpretation by an active agent can cancel transactions, alter databases, or send incorrect data to clients. Gartner estimates that an average Fortune 500 company could utilize over 150,000 AI agents by 2028, making individual real-time supervision mathematically impossible.

How to establish boundaries and accountability

Assigning blame to the software itself is legally and operationally impossible. Accountability must be established prior to deployment by clearly defining operational limits and human ownership.

• Map every agent: Maintain an inventory of all active systems, their data access permissions, and their allowed scope of execution.
• Define clear boundaries: Determine which tasks require mandatory human sign-off, such as final financial transactions or hiring decisions.
• Assign a human manager: Designate a specific professional responsible for auditing the logic, parameters, and results of each autonomous system.

Is your organization ready for agentic governance?

Relying solely on technological capabilities to define autonomy is a critical operational risk. Just because an agent can execute a complex workflow does not mean it should be authorized to do so without strict regulatory guardrails and human oversight frameworks.