AI agents are running HR processes, but the HRIS can't see who made the decision

AI agents are already making real personnel decisions, but HR systems still don't know who made them or why.

The invisible actor

When an agent rejects a résumé, approves a leave request or closes an employee's ticket, the HRIS records the fact. But who made the decision, under what authority and by what logic, the system does not know. The "author" field may show the name of the HR manager who launched the agent, or nothing at all. This creates a gap between how the formal org chart looks and how decisions are actually made.

What a ghost org chart is

A "ghost org chart" is the set of agents that de facto perform management functions but appear nowhere as entities with authority and accountability:

• A recruiting agent screens out 90% of candidates before a recruiter ever looks at them
• An approval agent approves or postpones leave according to set rules
• A support agent closes employee requests without escalating to a human
• An analytics agent produces the reports on which personnel decisions are based
• An onboarding agent guides a new hire on its own through the first weeks

Each of them affects real people's lives. None is listed in the staffing table or appears in the audit trail.

Why this creates risk

The problem is not the agents themselves. HRIS platforms were built for people and are not designed to treat agents as decision-making entities. An agent's traces are either lost or attributed to an employee who in fact decided nothing.

"The system stores the result but loses the decision's author, the

limits of their authority and the audit trail" — that is the heart of the problem.

This produces at least three categories of risk. Compliance: if an auditor asks why a particular candidate was rejected, there will be no answer. Fairness: if an agent systematically discriminates through proxy attributes, it is impossible to detect without a full decision log. Accountability: when something goes wrong, it is unclear who is responsible: the agent's developer, the manager who approved the parameters, or the leadership that adopted the tool.

What to change now

The authors of the piece propose several concrete steps. First, introduce agent identifiers in the HRIS: each agent should have a unique record describing its authority, constraints and owner. Second, log agent decisions separately from human actions, tied to the model version and the parameters in force at the moment of each decision. Third, formally delineate authority: define which decisions an agent makes on its own and which require a human, and codify this not only in prompts but in written policies.

What this means

Companies that deploy AI agents in HR without rebuilding their systems of record are building management on invisible infrastructure. Sooner or later that ends in an audit, a lawsuit, or simply the inability to explain why the system works the way it does. The time to think about agents' authority structure is now, before the ghosts become the majority in your org chart.