At a wealth management firm, the compliance officer showed me the logs from their AI-assisted anti-money laundering (AML) monitoring system. The firm's system flags unusual transaction patterns during client onboarding and ongoing reviews. Relationship managers can override with documented justification that gets logged for quarterly compliance review.

The compliance officer pulled a year's worth of records. Forty-seven overrides had been recorded. None had been reviewed. The relationship managers had learned which alerts were safe to dismiss. The system generated flags, and the compliance structure assumed someone was watching, but no one was.

In a recent piece, I wrote about quiet drift, the way board oversight of AI systems fades after deployment, while the model continues learning, adapting and making calls no one is watching closely enough to catch. The most common response I heard was a version of the same reassurance: we have humans in the loop.

That response deserves more scrutiny than it usually gets. Because in practice, the presence of human review and the existence of meaningful human oversight are not the same thing, and boards are routinely approving systems based on assurances that conflate the two.

Boards often spend months evaluating AI vendors, digging into security protocols and debating budgets. Once the system goes live, however, attention tends to shift elsewhere. The model continues operating as board oversight gradually fades from view.

But AI systems are not just another piece of infrastructure. In many cases, they can behave more like a senior executive who starts influencing thousands of decisions before anyone fully understands how those calls are being made. No board would hire an executive without expecting them to deliver clear decision rights, regular performance reviews, escalation protocols, team feedback and a documented trail of key decisions. Here's why AI shouldn't be treated any differently.

The author is Founder & President of the global consultancy Eunomia Risk Advisory. She writes: “Leaders must understand how power operates beneath the surface: how it’s signaled, challenged, and interpreted across borders and teams.” There is a discussion of the Power Distance Index (PDI), developed in the 1970s by Dutch social psychologist Geert Hofstede. She says that power dynamics “shape how organizations: launch new technologies and systems; handle ethical breaches and compliance failures; receive feedback—or fail to get it; and build or lose institutional trust.” She describes how “The GLOBE Project (Global Leadership and Organizational Behavior Effectiveness)-a multi-decade, cross-cultural study involving managers in over 60 countries, builds on Hofstede’s foundational work that Power Distance directly shapes leadership expectations.” She writes about how artificial intelligence/AI can play out in high and low-PDI cultures: “an AI tool can be ethically designed, but if it ignores cultural dynamics, it will fail functionally. Leadership must not only govern how AI is implemented, but also how its authority is communicated—and contested.” Finally, she discusses her tool “The POWER Scan^TM: A Framework for Culturally Fluent Leadership.”