New risk of AI leadership: more innovation, less human control

A new study on AI leadership reveals that artificial intelligence (AI) may be helping organizations produce more innovation while pushing human judgment, creativity and autonomy out of the process.

The study, Artificial Intelligence Leadership and the Decoupling of Human Agency: Evidence of Misaligned Innovation in Agentic AI Systems, published in Administrative Sciences, finds that AI leadership is linked to higher innovation activity while having a significant negative effect on human capabilities. Based on survey data from 3,079 respondents across industries and regions, the research warns that AI-driven innovation may increasingly operate independently of human agency, raising new risks for governance, accountability and long-term organizational control.

AI leadership is changing who drives innovation

In many organizations, AI systems now influence strategy, problem-solving, product development, workflow design and innovation decisions. The study argues that this shift is changing the function of leadership itself. AI leadership is no longer only about coordinating people and technology. It is increasingly about delegating decision-making authority to AI-enabled systems. This challenges a dominant assumption in management research: that AI strengthens innovation by improving human capability.

Much of the existing literature treats AI as an augmenting force, helping workers think faster, solve problems better and generate new ideas. The new study tests a more difficult possibility: that AI leadership can increase innovation outputs while reducing the role of human capabilities in producing them.

The research examines three core relationships: whether AI leadership increases innovation activity, whether AI leadership affects human capabilities, and whether human capabilities still explain innovation outcomes in AI-enabled organizations. Human capabilities in the study include autonomy, critical thinking, creative freedom, independent judgment, human expertise and the ability to challenge AI outputs. Innovation activity includes idea generation, experimentation, product development, innovation routines, team collaboration and AI-supported ideation.

The findings point to a structural shift. AI leadership had a strong positive effect on innovation activity, suggesting that organizations using AI-driven leadership practices can generate more innovation-related output. But the same AI leadership had a significant negative effect on human capabilities. In other words, the systems that helped organizations innovate also appeared to weaken the human skills and agency traditionally viewed as essential to innovation.

Notably, no significant link was found between human capabilities and innovation activity. There was no meaningful mediation pathway showing that AI leadership improves innovation through human capability development. Instead, innovation appears to emerge directly from AI-enabled leadership systems, not from stronger human involvement.

This pattern is described as structural decoupling. Innovation outputs continue to rise, but they become detached from human agency. The study argues that this should not be read as a simple efficiency gain. It may instead signal a deeper misalignment, where organizations produce more visible innovation while weakening the human judgment, creativity and oversight needed to keep those systems adaptable and accountable.

Innovation gains may mask capability erosion

Many organizations measure AI success through productivity, speed, output volume or innovation activity. However, the research suggests that such metrics may hide a quieter loss: the erosion of human agency inside innovation systems.

AI-driven leadership can make organizations faster. It can support rapid experimentation, automate idea evaluation, generate recommendations and structure innovation workflows. However, when employees increasingly rely on AI-generated outputs, their own role in questioning, interpreting and shaping those outputs may shrink. Over time, this can reduce autonomy, critical thinking and creative engagement.

The concern is not that human capabilities disappear overnight. It is that organizations may stop using them in meaningful ways. Employees may still be formally valued as creative contributors, but operationally sidelined by systems that define problems, generate options, rank priorities and recommend action. The study links this to institutional decoupling, where organizations continue to endorse human-centered values while relying in practice on AI-driven production systems. This gap creates governance risks.

If innovation is increasingly generated by AI systems, responsibility becomes harder to assign. Traditional management assumes that identifiable human decision-makers can be held accountable. In AI-driven innovation systems, decisions may emerge from complex interactions between algorithms, data inputs, system design and leadership structures. When outcomes fail or cause harm, it may be unclear who had authority, who approved the decision and who could have intervened.

The study also raises the issue of contestability. In healthy innovation systems, workers can challenge assumptions, question proposals and redirect flawed ideas. But in highly automated environments, AI-generated outputs may become harder to dispute. If employees do not understand how a recommendation was produced, or if the organization treats AI output as superior to human judgment, dissent can weaken.

This creates what the study describes as a risk of runaway technological optimization. AI systems may continue refining outputs according to internal performance metrics while drifting away from broader organizational values, human judgment or long-term adaptability. The result may be fast innovation that is also brittle, less transparent and harder to correct.

Human capabilities such as critical reflection, contextual judgment and creative synthesis are especially important in uncertain, non-routine and high-stakes situations. AI may be effective at accelerating known processes, but organizations still need people who can question whether the process itself is right. If AI leadership weakens those capabilities, short-term innovation gains may come at the cost of long-term adaptability.

The study’s large sample strengthens the warning. Respondents were drawn from multiple regions, industries and firm sizes, with strong representation from technology, finance, manufacturing, marketing, consulting, tourism and other sectors. Most participants had direct exposure to AI-enabled work environments, and many reported regular or daily AI use. This gives the analysis a broad view of how AI leadership is experienced across contemporary organizations.

The empirical results also showed strong measurement reliability and model fit. The analysis identified clear constructs for AI leadership, AI intensity and usage, human capabilities, and innovation activity. Structural modeling confirmed the core pattern: AI leadership increased innovation activity, reduced human capabilities, and showed no evidence that human capabilities mediated the innovation effect.

The authors describe the outcome as innovation without alignment. Organizations may become more innovative on the surface while becoming less human-centered underneath. The issue is not whether AI can produce outputs. It clearly can. The issue is whether those outputs remain linked to human agency, accountability and organizational learning.

Governance, not adoption, becomes the central challenge

The most important question is no longer whether organizations should adopt AI - it is whether they can govern AI systems that increasingly operate with reduced human involvement. This shift is especially urgent as AI systems become more agentic, systems that do not merely provide information but can generate, evaluate and execute decisions with limited human intervention. In such environments, leadership becomes a governance function that determines how much authority is delegated to AI and how much control remains with human actors.

The study introduces the AI–Human Alignment Diagnostic Scale as a proposed tool for assessing whether innovation systems remain aligned with human agency. The framework focuses on three dimensions: human agency integration, decision-making authority distribution, and governance visibility and control.

Human agency integration examines whether people remain actively involved in decision-making, evaluation and innovation processes. Decision-making authority distribution assesses whether AI systems are merely supporting human decisions or increasingly making and executing them. Governance visibility and control measures whether AI-driven decisions remain transparent, traceable, contestable and subject to oversight.

The framework distinguishes between aligned systems, partially decoupled systems and fully decoupled systems. A high-risk system is one where human involvement is low, AI authority is high and governance visibility is weak. Such a system may still produce strong innovation metrics, but those outputs can mask serious risks: declining human skill use, reduced accountability, limited ability to challenge AI decisions and weaker organizational control.

The study also identifies early warning signals of misalignment. These include rising reliance on AI-generated decisions without human validation, declining employee engagement in innovation tasks, fewer human overrides, limited employee understanding of AI-generated outcomes, reduced diversity of ideas and a growing sense that human input has little effect on final decisions.

AI adoption strategies should not focus only on deployment, output and efficiency. They should also protect human judgment inside innovation systems. Organizations may need mandatory human validation checkpoints, clear override procedures, audit mechanisms, review boards for high-impact AI decisions and designated roles responsible for monitoring human capability erosion.

The study also suggests that companies should assess AI systems periodically, not only when they are introduced. Because AI-driven systems evolve through repeated use, their effect on human agency may deepen over time. A system that begins as a tool can gradually become a decision architecture. Without active governance, workers may move from contributors to supervisors of systems they no longer fully understand or control.

The limitations of the research are also important. The study is cross-sectional, meaning it captures relationships at one point in time and cannot show how AI-driven decoupling evolves over years. It relies on survey-based perceptions rather than direct observation of algorithmic decision systems. It also does not fully control for differences in technological maturity, regulatory context or organizational culture across industries and countries.