I see this general statement often: “Governance becomes relevant the moment systems start influencing real decisions.”

It is a reasonable position. But it contains an assumption worth unpacking. Is the influence coming from the system? Or from the belief in the system?

That distinction sounds philosophical. It is entirely practical. But it’s also a comparison that most organizations tend to gloss over.

Two Modes of AI Decision-Making

There are two fundamentally different ways AI is involved in the decisions organizations are making right now.

The first: an AI system makes a decision. No human is in that equation. Humans deal with the outcomes. The credit is approved or denied. The content is moderated or published. The alert fires or stays silent. A human set up the system to make that call and walked away. The decision belongs to the machine.

The second: an AI system influences a human to make a decision. The human reads the recommendation. The human accepts it. The human acts on it.

Both modes are real. Both are happening at scale right now. And the response I received when I raised this was correct: both matter, and governance has to deal with the decisions that follow either way.

But the connective tissue between these two modes, the thing that makes both of them work and both of them dangerous, is the belief in the AI. The AI is the vehicle. The belief is the fuel.

The Mentormorphosis

In my book Infailible, I wrote about this belief structure, an ideology of AI. The gap between what AI can technically do and what people believe it can do. That gap runs about seven years, and it is widening faster than the capabilities are advancing. We have been, and remain, in a period in which belief is consistently more capable than technology.

What I am observing now is something I call the mentormorphosis. A transformation in how humans relate to AI as a source of authority. AI has become, for millions of people, a mentor they follow rather than a tool they evaluate. The recommendation arrives. The confidence is high. The interface is fluent and authoritative. The human accepts it.

People used to sit around at dinner having conversations about movies, until a question popped up, and someone would pull out their phone and say, “let me google it.” Now that same practice is happening with ChatGPT. But it’s much broader.

This is mainstream behavior. It is the operating mode of a significant portion of the modern knowledge workforce. Executives are making strategic decisions based on AI-generated analysis without auditing the underlying assumptions. Product teams are shipping features because AI research suggested the market wanted them. In some environments, clinicians are documenting AI recommendations without independent verification. Analysts are forwarding AI outputs as their own work product.

Even strong, knowledgeable people, fluent in a topic, are seeking second opinions, not through another human, but a Claude coworker.

The AI told them to. They believed it. They did it.

The governance implications of this are larger than those of an AI system making a decision independent of human involvement. An AI decision is visible. It is traceable. Someone set it up; someone can be held accountable. The mentormorphosis is invisible. The documentation never reflects that a decision was traced to an AI recommendation. The AI recommendation gets absorbed into the human decision, and accountability disappears with it.

The Belief That Systems Work

The first mode, AI making decisions autonomously, is also a belief problem, just a different one.

Organizations are setting AI systems to make consequential decisions on their own and walking away. The governance case for doing so is rarely examined. Because they believe the system works. The first hundred decisions looked right. The dashboard shows green. The output is plausible. The belief calcifies: this system is reliable; it has been running; nothing has broken visibly; we can trust it to keep running.

This is the belief in infallibility I have written about extensively. The quiet, incremental process by which human oversight decays in inverse proportion to the system’s apparent reliability. The decision to stop checking is never made consciously. They just gradually stopped because the checks kept confirming what the system was producing, and at some point, the confirmation became the assumption.

When the assumption is wrong, the system has been making unchecked decisions for months. The accountability chain runs back to the human who set it up and the humans who stopped watching. The belief that the system was working was the governance failure. The system simply ran.

Does Belief Change Who Is Accountable?

This is where the philosophical question becomes a practical one. Perception versus reality is a real governance challenge. But does it change who is responsible when something goes wrong?

No.

Who set up the AI to make decisions on its own? That person is responsible. The scope it was given, and the oversight that was or wasn’t applied.

Who in the organization made a decision based on what AI recommended? That person is responsible. The fact that AI was in the reasoning chain fails to transfer accountability away from the human who acted.

“The AI made me do it” holds no weight as a legal defense, no standing as an ethical defense. It is a description of a situation in which a human followed a recommendation without independent evaluation, which is itself a human decision.

The second case is admittedly more complicated in practice. If a bad decision traces to an AI recommendation, the organization is unlikely to surface that fact voluntarily. No one wants to be the executive who explains to the board that a major strategic error resulted from following AI output without verifying it. The AI recommendation disappears from the decision documentation. The human will present the decision as their own analysis. The accountability will appear clean even though the process was compromised.

This is where the governance challenge is actually hardest. The AI decision cases are tractable — at least the system is identifiable, and the setup decision is traceable. But in the mentormorphosis cases, a human’s AI-dependent decision-making is invisible to every governance mechanism, audit, and accountability structure in the organization.

The EU AI Act and the Belief Gap

I have argued that the EU AI Act is primarily positioned for a version of AI that has yet to fully arrive in production. The Act was written in significant part based on the belief about what AI will become, rather than a granular analysis of what it is doing today in most enterprise deployments.

This is a softer critique than it sounds. The belief gap is a legitimate input to governance. If organizations and regulators believe AI has certain capabilities, those beliefs drive behavior. And behavior driven by the belief in AI capability is just as real in its consequences as behavior driven by actual AI capability.

But it does produce governance frameworks calibrated to imagined threats. The regulatory requirements for high-risk AI systems are, in many cases, based on scenarios that represent a small fraction of actual AI deployments. The vast majority of AI in production today is software that needs to be governed using the governance frameworks organizations already have in place. The AI label has made that governance feel insufficient, which has produced a regulatory and market response calibrated for something more dangerous than what is mostly running.

The belief that AI is more capable and more autonomous than it currently is produced governance requirements designed for that more capable system. Organizations are now trying to comply with those requirements, even for systems that fail to meet the assumptions. The compliance burden is real. The threat it addresses is partially imaginary.

The Governance of Belief

The practical implication of all of this is that the hardest AI governance problem is human, far more than it is technical.

Governing an AI system that makes autonomous decisions is tractable. You identify the system. You verify the identity. You evaluate the actions. You maintain the audit trail. You trace accountability to the human who deployed it.

Governing the organizational culture that has decided AI recommendations are authoritative is much harder. You cannot audit a belief. You cannot hash-chain the reasoning process by which an executive decided to follow an AI-generated strategy. You cannot independently verify whether a human’s stated reasoning was actually their reasoning or a rationalization of what the AI told them to do.

What you can do is require transparency. Require that AI inputs to significant decisions be documented. Require that the governance chain include the AI system as an input, alongside the human as the decision-maker. Require that recommendations from AI systems be independently evaluated before consequential action, rather than accepted on the authority of the model’s confidence.

These are governance requirements. They are uncomfortable because they require organizations to formally acknowledge the degree to which their decisions are AI-influenced. And that acknowledgment changes the accountability picture in ways that most organizations would prefer to avoid.

The belief in AI is the bigger governance issue. The belief is wrong about a great many things, but that is secondary. But because unchallenged belief produces unchecked influence, and unchecked influence without accountability is exactly the governance failure that every framework in this space is supposed to prevent.

Whether a human believes AI is making decisions on its own or believes AI recommendations are authoritative and follows them, the accountable party is the same.

A human.

Who set it up? Who believed it? Who acted on it?

The belief fails to transfer the accountability. It just makes it harder to find.

---

If you find this content valuable, please share it with your network.

Follow me for daily insights.

Book me to speak at your next event.

Start managing your agents for free.

Chris Hood is an AI strategist and author of the #1 Amazon Best Seller Infailible and Customer Transformation, and has been recognized as one of the Top 30 Global Gurus for Customer Experience. His latest book, Unmapping Customer Journeys, will be published in 2026.