When AI Chooses

Key Takeaways

• Delegated Authority: As AI systems become more capable, boards will need to decide not only what AI can do reliably, but which decisions they are willing to let it make.
• Human Oversight: Growing familiarity with autonomous AI does not necessarily mean people will accept its use for decisions with significant consequences for employees, communities and other stakeholders.
• Accountability Remains Human: When AI-made decisions are perceived as harmful or unfair, responsibility is likely to fall on the people and organizations that authorized the system to act.
• The Human Threshold: Boards may need to identify the point at which a decision becomes consequential enough that stakeholders still expect a human to make it.
• A Moving Boundary: The division of decision-making authority between humans and AI will change as technology, controls and stakeholder expectations evolve.

Deep Dive

Imagine the boardroom in 2036. Artificial intelligence has become part of how the company operates. It tests capital choices, monitors risk, compares strategic alternatives and makes thousands of decisions within limits established by management and the board.

Over a decade, those limits have expanded. Analysis became recommendation; recommendation became delegated action. Each expansion was supported by successful decisions and controls that continued to work. The company became faster and more profitable.

By then, people have also spent years becoming accustomed to AI acting on their behalf, from autonomous transportation to everyday financial and consumer decisions. Inside the company, each expansion is tested against performance, risk and control. Outside it, greater autonomy draws little notice as long as it works.

That direction is already visible today. Governance guidance increasingly focuses on how boards should allocate decision rights between humans and autonomous systems, establish escalation points and determine where human judgment should remain. World Economic Forum; Directors & Boards.

Then Comes a Different Choice

The company needs to reduce manufacturing capacity by 15 percent. That decision remains human. Management recommends the restructuring, and the board approves its strategic and financial parameters. What directors do not decide is exactly where the reduction should occur.

By 2036, the company’s autonomous planning system has long been authorized to choose among operating alternatives within approved limits. It compares dozens of configurations, weighing investment requirements, transportation costs, customer disruption, labor availability, employee impact, community effects and likely public reaction.

Its conclusion is economically compelling: close four plants, eliminating 8,000 jobs.

Management carries out the restructuring. The AI has not malfunctioned, and the controls have not failed. The system has done what humans authorized it to do. It has simply made a choice that, ten years earlier, almost certainly would have been human.

Who Chose These Plants?

The reaction starts where the plants are closing. Employees who expected to retire there talk to local reporters. Families worry about mortgages, local businesses about lost income and mayors about what happens next.

Then reporters ask how the four sites were selected. The company does not hide the answer. Management and the board approved the restructuring; the autonomous system selected the facilities under the company’s existing decision framework.

The issue spreads beyond the affected communities. National coverage follows. Politicians raise questions. Employees elsewhere wonder whether their locations could be next. Investors ask whether the company underestimated the consequences of allowing an autonomous system to make the selection.

The system may even have predicted the reaction accurately. The issue is not whether AI understood the consequences. It is whether people accepted its authority to choose them.

No human selected those four plants, although executives and directors remain responsible for the framework that allowed the system to do so. For the first time, the company is discovering that confidence in what AI can do may not be the same as acceptance of what it should be allowed to decide.

How Did the Line Move?

Probably gradually. EY’s 2026 global AI sentiment research surveyed 18,152 people across 23 markets. Eighty-four percent had recently used AI, while 16 percent had used systems capable of acting without human intervention. Yet 66 percent still said human oversight remained essential. EY, 2026.

That combination does not tell us that people will someday accept AI selecting plant closures. It suggests something narrower: use and comfort with autonomy do not necessarily move at the same speed.

Familiarity and context matter too. A 2025 preregistered AI & Society experiment involving 2,100 adults in Hungary found that decisions involving human participation were generally more trusted than AI-supported decisions in medical diagnosis, hiring and transportation; financial investing was the exception. The researchers also found that greater understanding of AI could reduce some of the negative effect on trust, while trust varied depending on the type of decision involved.

When harm occurs, responsibility may move in the other direction. A 2023 CHI study by Gabriel Lima, Nina Grgić-Hlača and Meeyoung Cha found across three experiments involving 1,153 participants that perceptions of unfairness and harm increased blame toward the humans who designed and used AI systems, while having much less effect on blame directed toward the machines themselves. CHI 2023.

Together, the research points to a tension. Familiarity can make greater delegation easier to accept. But when an AI-made choice produces consequences people experience as unfair or deeply human, they may still look back toward the people who gave the system its authority.

The Human Threshold

Boards are already being encouraged to ask what AI can do, what controls it requires and where human judgment should remain. Directors & Boards. The harder future problem may be that the line is not theirs alone to draw.

The point at which an AI-made choice becomes consequential enough that stakeholders still expect a human to make it may become the board’s human threshold. That threshold is unlikely to stay in one place. Successful experience may move it outward; perceived unfairness or serious human consequences may pull it back. Investors, employees, regulators and the broader public may also draw the line differently.

Nor is this simply reputational risk. Stakeholders might accept the economic outcome while objecting to the fact that this particular choice was delegated at all.

The GRC Report is your premier destination for the latest in governance, risk, and compliance news. As your reliable source for comprehensive coverage, we ensure you stay informed and ready to navigate the dynamic landscape of GRC. Beyond being a news source, the GRC Report represents a thriving community of professionals who, like you, are dedicated to GRC excellence. Explore our insightful articles and breaking news, and actively participate in the conversation to enhance your GRC journey.