The AI Trust Paradox: Nick Curum Questions Executive Reliance on Unreviewed AI Output

In a recent LinkedIn post, Nick Curum highlights a significant disconnect in how executives perceive and trust Artificial Intelligence (AI) output, particularly within finance, risk, and sustainability functions. Curum raises critical questions about the unverified reliance on AI-generated information, citing a Workiva survey that reveals a stark contrast between trust in AI output and trust in the underlying data.

According to the survey conducted in August 2026 among 2,272 professionals, a staggering 84% of executives reported trusting AI output without review. However, only 11% expressed the same level of confidence in the data that feeds these AI systems. Curum emphasizes the inherent contradiction in these figures, stating:

Those two numbers cannot both be reasonable.

Curum’s analysis suggests that this discrepancy points to a potentially dangerous oversight in decision-making processes. He elaborates on the implications:

Either the data is good enough to act on, or the output built from it needs a human before it leaves the building.

Questioning the Foundation of AI Trust

The core of Nick Curum’s argument centers on the principle of calibration, drawing an analogy from process engineering. He posits that confidence in any reading or output is fundamentally limited by the confidence one has in the source data or sensor. In reporting and AI-driven decision-making, Curum argues that this principle is being inverted.

Adding another layer to his concern, Curum points out that the same survey revealed a troubling statistic: 26% of the surveyed executives admitted that an internal audit had already identified AI errors that had made their way to the board or external audiences. This indicates that the issue is not merely theoretical but has real-world consequences.

The Need for Human Oversight and Data Integrity

While acknowledging that Workiva, the survey’s sponsor, operates in the audit and reporting software space, Curum urges readers to look beyond potential framing. He identifies two straightforward decisions that can help bridge the gap between AI output and data integrity:

• Naming the specific data sources that the AI is permitted to draw upon.
• Identifying the human responsible for the final output that is presented.

Curum suggests that implementing these measures is not an insurmountable project, estimating that each could be addressed within an afternoon. He believes these steps are crucial for ensuring accountability and maintaining trust in AI-assisted business processes.

Nick Curum, who writes for individuals making critical capital decisions under pressure, concludes by urging professionals to carefully consider the reliability of their AI systems. His post serves as a vital reminder that while AI offers powerful capabilities, the integrity of its foundational data and the necessity of human review remain paramount.

In process engineering, confidence in a reading is capped by confidence in the calibration. You never trust the instrument more than you trust the sensor behind it.

Curum’s insights underscore the importance of a balanced approach to AI adoption, one that leverages technology while upholding rigorous standards for data governance and human accountability.