What is named accountability?

Named accountability means a specific individual, identified by name, answers for a decision, a system or an outcome. Not a team, a committee or "the business": one person who can be asked what happened and why, and who carries the consequences. It is the opposite of diffuse responsibility, where everyone was involved and nobody decided.

Why does it matter when agents act?

Agents make accountability easy to lose. An agent drafts, a reviewer approves, a manager set the limit, a builder wrote the brief, a vendor supplied the model. When something goes wrong, each can point to another. Named accountability cuts through that by deciding in advance who answers for what.

Regulators expect it. The UK's Senior Managers and Certification Regime is built on named individual accountability, and frameworks such as the NIST AI Risk Management Framework put roles and responsibilities at the heart of their Govern function; see the NIST Govern function and accountability for agents.

How does it work in practice?

For each agent, the organisation names:

• an agent owner, who answers for the agent in production: its scope, depth, certification and performance;
• named approvers, who release specific held actions within their authority;
• a builder, who answers for the brief, sources and bounds as designed;
• a senior manager, who answers for the area the agent works in.

Each name sits against a defined scope, and each approval is filed in the record under the approver's name. A held action waits for a named person; it never waits for "someone in finance". The mechanics of setting this up are in how to set up named approvers for an AI agent, with a ready format in the named approver roster template.

What is an example?

A wholesale distributor's agent drafts supplier payments and holds anything above the customer's limit. The roster names two finance officers as approvers, each with a stated limit and a current certificate. One approves a payment to a changed account on the strength of an emailed request. When the error surfaces, the record shows who approved, what evidence they saw and which credential they held. The question is not "how did the system let this through?" but "why did this approver accept external evidence?", and that question leads to a specific training fix.

How is it different from AI accountability?

AI accountability is the broad principle that organisations must answer for the AI they use: for its outcomes, its fairness, its compliance. Named accountability is one way of making that real, by assigning each part of the answer to a person. An organisation can claim AI accountability in a policy without naming anyone. It cannot claim named accountability that way.

There is also a narrower term, the named approver: the person who releases a specific held action. Every named approver is an instance of named accountability, but accountability also covers owners, builders and senior managers who never touch the queue.

Which terms are related?

• Agent owner: the person accountable for one agent.
• RACI matrix: a common way to record who is accountable for what.
• Separation of duties: making sure one person does not both build and approve.

For the wider topic, read AI literacy and certification for people who work with agents.