The Human Authority Gap
AI Is Gaining More Power at Work. Leadership Has to Catch Up.
By Dr. Marisol Capellan
October 4, 2026
Artificial intelligence is rapidly changing what organizations can automate, accelerate, and delegate. Yet one of the most consequential questions about AI at work has received far less attention than productivity, adoption, or technical capability: What happens to human authority when intelligent systems begin participating in consequential organizational decisions?
For years, the dominant workplace conversation about artificial intelligence has centered on what the technology can do. Can it draft the report, analyze the data, identify risk, screen applicants, answer customers, predict demand, or recommend the next action? Those questions remain important, but they are becoming incomplete. As AI systems move from generating information toward recommending, prioritizing, coordinating, and increasingly executing organizational actions, leaders face a more difficult question. They must decide not only how AI should be used, but what authority it should be allowed to exercise and what authority humans must continue to possess.
That distinction may become one of the defining leadership challenges of the AI era.
Imagine a manager reviewing a recommendation generated by an AI system. The system has already analyzed the available information, ranked possible outcomes, and identified what it considers the best course of action. The manager reads the recommendation and approves it. Another system then prioritizes the action, and an automated workflow executes the decision. On paper, a human made the final call.
But that description leaves out nearly everything that matters.
Did the manager understand how the recommendation was produced? Did she have access to information that might contradict it? Did she possess the expertise and time required to independently evaluate it? Could she reject the recommendation without creating friction or having to explain why she had departed from the system’s supposedly objective analysis? Would her organization support that decision? If the recommendation ultimately harmed an employee, applicant, customer, patient, or other stakeholder, could she reverse what happened?
A human may remain in the process while possessing surprisingly little meaningful authority over it.
That is the problem I describe as the Human Authority Gap.
When Responsibility and Authority Begin to Separate
The Human Authority Gap is the organizational distance between the consequential influence or practical authority exercised through artificial intelligence and the human authority, judgment, accountability, voice, and capacity to intervene that are necessary to govern its consequences responsibly.
In simpler terms, the gap begins to appear when people remain responsible for outcomes while losing some of the practical authority necessary to shape them.
This matters because organizations frequently respond to concerns about automated decision-making with a reassuring statement: A human is still in the loop. The phrase sounds protective, but it can conceal more than it reveals. Human participation is not the same thing as human authority, just as human approval is not necessarily evidence of independent human judgment.
A manager can click “approve” without meaningfully evaluating a recommendation. An employee can technically have the right to appeal a decision while having no realistic access to anyone empowered to reverse it. A professional can be given an override button while working in a culture where departing from the algorithmic recommendation is implicitly discouraged. A leader can formally remain accountable for an outcome while depending almost entirely on a system whose reasoning she did not independently test.
The deeper question is therefore not whether a human appeared somewhere in the workflow. It is whether that human had enough information, discretion, competence, independence, time, authority, and organizational support to meaningfully affect the outcome.
Research has been moving toward this problem for years. Scholars studying autonomous systems have long examined what happens when traditional assumptions about responsibility break down because people no longer possess complete knowledge or control over technological systems. Matthias’s work on the “responsibility gap” dates to 2004, while later research on meaningful human control has explored what substantive human involvement should require rather than treating human presence as sufficient by itself (Matthias, 2004; Santoni de Sio & Mecacci, 2021).
The current wave of artificial intelligence makes those questions newly urgent for leadership. Emerging research is examining authority allocation between humans and AI, meaningful human oversight, decision rights, institutional authority, and the risk that leaders can become increasingly dependent on AI-generated reasoning while formally retaining responsibility for the decisions that follow (Landers, 2026; Sebastian et al., 2026; Zhu et al., 2026).
The issue is no longer theoretical. Organizations are actively deciding, often without describing the decision in these terms, how much practical authority intelligent systems will possess.
Leadership Has Always Been About More Than Influence
My interest in this problem did not begin with artificial intelligence. My earlier work has been organized around a proposition that remains central to how I understand leadership: leadership is a responsibility.
Leadership is often described through influence, vision, motivation, strategy, or performance. But leadership also involves power. Leaders make decisions that influence other people’s work, opportunities, resources, careers, and futures. The ability to exercise consequential authority creates a corresponding obligation for how that authority is used.
AI complicates this relationship because the leader formally responsible for an outcome may no longer be the only meaningful source of influence over how the decision was produced. Judgment may be distributed across managers, organizational policies, datasets, algorithms, software vendors, automated workflows, and intelligent agents. Responsibility can remain human even as some of the practical authority generating the outcome begins to move elsewhere.
That is why the leadership question cannot simply be, How should leaders use AI? A more consequential question is emerging: How should leaders exercise authority responsibly when the systems through which authority is exercised are no longer entirely human?
That question is the foundation of what I call Responsible Authority™.
Responsible Authority is a leadership and organizational framework for the disciplined design and exercise of consequential power so that authority remains meaningfully connected to responsibility, independent judgment, voice, accountability, and the capacity to intervene and repair outcomes.
The important word is not simply “responsible.” It is authority.
A technically sophisticated and ethically designed AI system cannot decide for an organization who should be allowed to use it, where its authority should end, which decisions require human judgment, who may challenge its recommendations, or who must repair an outcome when something goes wrong. Those are questions of leadership and organizational design.
Five Questions Every Authority System Must Answer
Responsible Authority begins with five connected questions. The first concerns authority itself: Who actually has the power to decide or act? The answer may not be the person listed on the organizational chart. If an AI system determines which information reaches a manager, which candidates appear first, which risks receive attention, or which option becomes the default recommendation, it can exercise substantial practical influence even without possessing formal authority.
The second concerns responsibility: Who owns the consequences? Organizations may continue assigning accountability to managers, executives, professionals, boards, or institutions, but that accountability becomes less meaningful when those actors do not possess adequate control over the processes generating the outcome. Saying that “the human remains responsible” does not solve the problem if the human cannot independently evaluate, reject, suspend, escalate, or repair the decision.
The third is judgment, and it may prove especially important. Artificial intelligence can produce recommendations with extraordinary speed, fluency, and apparent confidence. That capability is valuable, but it can also change human behavior. A manager may gradually move from asking, What do I think? to asking only, Does the AI’s answer seem plausible? Those are very different forms of cognition. Meaningful human judgment requires the ability and willingness to independently evaluate, disagree, and alter the outcome.
The fourth is voice. Authority is experienced not only by the people who exercise it but by the people living with its consequences. If an employee believes an AI-supported performance evaluation is wrong, if an applicant disputes an automated classification, or if a manager believes a system-generated recommendation is inappropriate, can that concern reach someone with the power to reconsider the decision? Research on employee voice and algorithmic decision-making suggests that fairness and trust can be strongly affected by whether people believe they can question and contest consequential decisions (Morrison, 2014; Moritz et al., 2026).
The fifth is repair. Organizations spend enormous effort trying to prevent failure, but responsible authority must also account for what happens after failure occurs. Who can stop continuing harm? Who can reverse a decision? Who can correct the information? Who can restore an opportunity? Who can modify the system or escalation process so the same problem does not recur? An organization can identify who was technically accountable and still lack anyone with sufficient authority to make the situation right.
Taken together, these questions move the conversation beyond whether AI is trustworthy or compliant. They ask whether the organizational system of authority surrounding AI remains coherent.
Leadership Is Becoming the Stewardship of Authority
If authority becomes more distributed, the leadership task changes with it.
I use the term Leadership Stewardship to describe leaders’ responsibility for deliberately designing, monitoring, and recalibrating how authority is distributed across people and intelligent systems. Stewardship itself is not a new leadership concept. It has long been associated with responsibility for institutions, stakeholders, and the long-term consequences of decisions (Hernandez, 2008). What changes in an AI-mediated organization is the object being stewarded.
Leaders increasingly become stewards of authority itself.
They must determine which decisions AI may support, which it may execute, where mandatory human review belongs, what authority humans must retain, when an AI-enabled process can be stopped, who can override it, how disagreement will be handled, and how those boundaries should change as systems become more capable.
This creates an important leadership shift. AI may reduce the number of decisions leaders make personally while increasing their responsibility for designing the systems through which decisions are made.
Leadership is not disappearing. Its object is changing.
The future executive may spend less time personally producing every judgment and more time defining who or what is authorized to make consequential judgments in the first place.
The Next Question Is Not Whether Humans Have Authority, but Whether They Have Enough
This leads to what may become one of the most important empirical questions in the field: How much human authority is sufficient?
A manager can technically possess final decision rights while lacking meaningful authority because she does not have enough information, expertise, discretion, time, or institutional protection to use them. Another manager may delegate substantial routine activity to AI while retaining meaningful authority because she has visibility into the process, access to independent evidence, real override power, clear escalation routes, and the ability to correct outcomes.
The relevant question is therefore not whether a decision is “human” or “AI.” The more useful question is whether human authority remains adequate for the responsibility attached to it.
That adequacy will almost certainly depend on the stakes. An automated scheduling recommendation does not require the same authority architecture as a decision involving termination, promotion, credit, healthcare, safety, or another person’s access to a consequential opportunity. As risk and consequence increase, the human authority required to make responsibility meaningful may need to increase as well.
This is where the discussion needs to move from philosophy into research. We need to understand what combinations of decision rights, information access, expertise, AI literacy, discretion, override capacity, escalation rights, institutional protection, and repair capability make human accountability substantive rather than symbolic.
Those questions should be measured, tested, and challenged with evidence.
Why This Is Different From AI Governance
Organizations are already investing heavily in responsible AI and AI governance, and that work is essential. Responsible AI typically addresses whether systems are designed and deployed consistently with principles such as fairness, safety, privacy, transparency, accountability, and reliability. AI governance addresses the policies, controls, standards, documentation, roles, and risk-management structures surrounding those systems.
Responsible Authority addresses a different layer of the problem.
An organization can have a sophisticated AI governance program and still struggle to answer a very basic set of leadership questions. Who actually made this decision? Who could have stopped it? Who was required to think independently? Who could challenge it? Who could reverse it? Who could repair the consequences?
AI governance asks how artificial intelligence should be governed.
Responsible Authority asks how organizational authority should be governed when artificial intelligence becomes part of consequential action.
That distinction may become increasingly important because technology can be compliant without the surrounding authority structure being coherent.
AI May Increase Leadership Power While Reducing Direct Control
There is also a paradox embedded in this transformation.
Artificial intelligence can dramatically expand a leader’s reach. An executive supported by intelligent agents may oversee more transactions, workflows, decisions, customers, or employees than would have been possible under traditional management structures. But that same executive may possess less direct visibility into the thousands of micro-decisions through which those outcomes are produced.
In other words, AI can increase leadership power while decreasing direct leadership control.
That is not necessarily a reason to avoid delegation. Organizations could not scale if leaders personally made every operational decision. Delegation is fundamental to management. But AI introduces a form of delegation with important differences. Human subordinates can explain their reasoning, participate in relationships of trust, develop judgment, receive coaching, interpret social context, and assume organizational responsibility in ways that artificial systems cannot.
The challenge is therefore not whether leaders should delegate to AI.
The challenge is how leadership responsibility changes when the recipient of delegated authority can act consequentially without participating in human systems of responsibility in the same way.
This is leadership under delegated authority.
Building the Research Program
Responsible Authority should not become another attractive management phrase detached from evidence. Its usefulness will depend on whether it produces research capable of explaining what actually happens inside organizations.
That is why I am developing Human Authority at Work, a research agenda focused on how authority, responsibility, managerial judgment, employee voice, accountability, trust, and leadership legitimacy change as artificial intelligence becomes increasingly embedded in organizational decision-making.
The central scholarly question is straightforward: How should leaders exercise and preserve responsible human authority when consequential organizational decisions and actions are increasingly shared with artificial intelligence?
That question opens a series of empirical problems. What happens when leaders remain accountable while practical authority declines? Under what conditions does AI improve managerial judgment, and under what conditions does it replace it? How much override capacity is required before human oversight becomes substantive? Which decisions should retain mandatory human authority? Does repeated delegation of judgment-intensive work eventually weaken managerial expertise? Do people perceive AI-mediated authority as more legitimate when meaningful appeal mechanisms exist? And how do different cultures’ assumptions about hierarchy, autonomy, authority, and responsibility shape acceptable patterns of human-AI delegation?
These are fundamentally leadership and organizational-behavior questions, even though they intersect with AI governance, information systems, human-computer interaction, business ethics, and the future of work.
The field is not empty. It is forming.
That is exactly why leadership scholarship should enter the conversation now, before technological practices harden into organizational assumptions that were never deliberately examined.
The Leadership Question Organizations Should Ask Now
Organizations do not need to wait for a decade of research before examining their own authority systems. Leaders can already ask whether they know what AI has been authorized to influence or execute, who decided to delegate that authority, who remains accountable, whether that person possesses enough practical power to justify the responsibility, where independent human judgment is mandatory, who can override the process, how an affected person can challenge a decision, and who can repair an erroneous outcome.
They should also ask a harder question: How will we know if human judgment is slowly disappearing before anyone consciously decides to give it up?
An organization can possess advanced AI tools, a detailed AI strategy, and formal governance policies while still lacking a coherent leadership authority strategy.
That gap may prove increasingly consequential.
Artificial intelligence is forcing leadership to reconsider concepts that once appeared comparatively stable: authority, responsibility, judgment, voice, accountability, and control. For much of management history, consequential organizational power ultimately flowed through human actors. Technology could inform, constrain, automate, or amplify decision-making, but humans remained relatively visible at the center of authority.
That assumption is becoming harder to sustain.
AI increasingly participates in how organizations see problems, generate options, rank people, allocate resources, evaluate performance, coordinate work, recommend decisions, and execute actions. The defining leadership question may therefore not be whether artificial intelligence will become a leader.
It may be what becomes of leadership when authority itself becomes distributed.
Responsible Authority begins with the proposition that responsibility does not disappear when authority is delegated. If anything, the obligation becomes more demanding. Leaders must determine what authority can be delegated, what must remain human, where independent judgment cannot be surrendered, who can challenge the exercise of power, when intervention is required, and who possesses both the responsibility and ability to repair failure.
The future of leadership may depend less on whether leaders personally make every consequential decision and more on whether they can responsibly design and govern the systems through which consequential decisions are made.
That is the work of Leadership Stewardship.
And it leaves organizations with a deceptively simple question they should be able to answer before AI becomes even more autonomous:
When AI acts, who remains responsible?
About the Research
This essay is based on Responsible Authority™ Research Note No. 1, The Human Authority Gap: Who Decides, Who Can Override, and Who Remains Responsible When AI Acts at Work? The full conceptual research note situates Responsible Authority within scholarship on responsibility gaps, meaningful human control, ethical and responsible leadership, stewardship, employee voice, human oversight, and AI decision rights.
Permanent DOI: https://doi.org/10.5281/zenodo.23149353
Suggested Citation
Capellan, M. (2026). The human authority gap: Who decides, who can override, and who remains responsible when AI acts at work? (Responsible Authority™ Research Note No. 1). The Capellan Institute. https://doi.org/10.5281/zenodo.23149353
About Dr. Marisol Capellan
Dr. Marisol Capellan is a leadership researcher, author, executive leadership coach, and founder of The Capellan Institute. Her work examines leadership responsibility, authority, organizational systems, managerial judgment, and the changing relationship between human decision-making and artificial intelligence. Her earlier work, including Leadership Is a Responsibility, centers on the proposition that leadership power creates corresponding obligations. Her current research extends that inquiry into AI-mediated organizations through the development of Responsible Authority™ and the Human Authority at Work research agenda.
For research collaborations, executive briefings, speaking, or media inquiries, contact Dr. Marisol Capellan at coaching@marisolcapellan.com