September 25, 2026, (Inside AI) — Companies are pouring billions into artificial intelligence, yet many struggle to show a return. The problem may not be the technology. It may be the people who quietly refuse to use it. Employees are overriding AI systems even when those systems work, says Harvard Business School professor Das Narayandas. He argues that this is not resistance but self-preservation.
"When accountability remains personal," Narayandas says, "people will take steps to protect themselves."
Narayandas made the argument in a forthcoming Harvard Business Review article, an early look at which was shared with subscribers. The piece will appear in the November-December issue.
The core issue is decision rights. When an AI system is deployed, it often shifts authority away from individuals. Those individuals may lose discretion over decisions they once controlled. If the system fails, they may still be blamed. So they override it, treating AI insights as mere suggestions.
Narayandas points to a global bank that approved an AI system to automate routine credit decisions. The system had proven itself in a pilot: it processed applications faster, reduced errors, and promised cost savings. Risk management and compliance signed off. But six months after launch, humans were still making most credit decisions.
This pattern is widespread. A 2025 survey by McKinsey found that only 16% of companies reported significant bottom-line impact from AI. Another study by MIT researchers in 2024 showed that 85% of AI projects fail to deliver on their promises. The reasons often involve people, not algorithms.
Narayandas says leaders must answer four questions before deploying AI: Who loses discretion if the system works? Who gains authority and over which decisions? Who explains outcomes when the system is followed? Who absorbs transition risk during the learning curve?
If these questions are not answered explicitly, they will be answered implicitly by individual operators protecting their turf. The result is a shadow system where AI is nominally in charge but humans still pull the levers.
"Full adoption will occur only when approving a system's decisions does not feel like a professional self-sacrifice," Narayandas says.
The solution is not to remove human judgment but to move it to higher-value tasks: assessing portfolio risk, monitoring model drift, and handling edge cases. Organizations must provide clear guardrails and tools that help employees explain AI decisions. Leaders must publicly absorb errors as part of the learning process.
This is not just a management challenge. It is a cultural one. Companies that treat AI as a tool for augmentation, not replacement, may see better results. Those that ignore the human cost of lost authority will continue to see overrides.
Read: AI in Leadership Communication: Balancing Efficiency and Authenticity
In related news, Harvard Business Review is hosting a community of practice on AI adoption. The final session on September 29 will feature Michael Lurie, chief transformation officer at Bayer, discussing how AI is transforming leadership.
