The Problem with AI Was Never Intelligence, It Was Authority
- Mar 11
- 6 min read
Updated: Mar 19
Written by Steve Butler, CEO of Luminary Diagnostics
Creator of Butler's Six Laws of Epistemic Opposition and the Seven Laws of Agentic Safety – together forming the constitutional AI safety framework. He is the CEO of Luminary Diagnostics.
For months now, we have been circling an uncomfortable truth about artificial intelligence. The systems reshaping our organisations, markets, and institutions are not failing because they lack intelligence. They are failing because we never decided who is actually in charge.
My November 2025 article described the hidden crisis inside modern AI, systems that produce confident answers faster than humans can properly evaluate them. The second article explained why walking away was impossible until I understood what that meant.
This third piece is not a product announcement. It is not a technical explanation. It is a reframing. Once you see this clearly, it becomes very difficult to see AI in the same way again.
The category error we keep making
Nearly every serious conversation about AI safety, governance, or risk starts in the wrong place.
We talk about accuracy. We debate hallucinations. We argue about bias, guardrails, and better training data. We publish policies. We create committees. We reassure ourselves that the problems are being handled.
All of that activity feels responsible. None of it reaches the root. The deeper failure is simpler, and far more unsettling. We have built systems that influence real decisions without ever making authority explicit.
When an AI system recommends an action, prioritises an option, or predicts an outcome, something subtle happens. Humans begin to treat the output as if it carries authority, even when no one has consciously granted it.
The suggestion feels neutral. The confidence feels earned. The speed feels helpful. And yet something critical has already shifted.
Responsibility begins to blur. Confidence inflates without ownership. And when something goes wrong, a basic question becomes surprisingly hard to answer.
Who actually decided this? That question is not philosophical. It is operational. It sits underneath accountability, liability, ethics, and trust. And right now, most AI systems cannot answer it. That is not a technical failure. It is a structural one.
Why intelligence alone was never the risk
It is tempting to believe that the danger of AI lies in how intelligent it becomes. Smarter systems feel more threatening. They outperform us. They surprise us. They make fewer obvious mistakes. But intelligence, by itself, has never been the thing that caused harm.
Human history is full of intelligent tools, calculators, engines, markets, algorithms. None of them became dangerous simply because they were clever. They became dangerous when authority was unclear, when incentives misaligned, or when responsibility could be plausibly denied.
AI is no different. The risk is not that systems are intelligent. The risk is that intelligence is now arriving without friction, without hesitation, and without an obvious place for responsibility to land.
How speed quietly breaks responsibility
For most of human history, intelligence moved at human speed. People argued. They hesitated. They disagreed. Decisions took time, and time created friction. That friction mattered. It forced ownership. It preserved accountability.
Artificial intelligence removes that friction. Intelligence now arrives instantly, confidently, and at scale. When systems operate faster than human reflection, authority does not disappear. It migrates.
Not because anyone chose it to move, but because nothing in the system prevented it from doing so. This is why organisations rarely fail suddenly. They fail gradually, while everything still appears to be working. Warnings are raised. Concerns are noted. And momentum carries on regardless.
By the time harm becomes visible, the decision has already hardened into reality, and responsibility has already diffused beyond reach. We have seen this pattern before.
Major failures rarely begin with catastrophe. They begin with systems that continue to operate normally while authority becomes diffuse, warning signals become muted, and the consequences of error grow larger than anyone realises.
The Challenger launch decision, the global financial crisis in 2008, and more recently the Boeing 737 MAX failures all revealed the same underlying structure, responsibility blurred, signals distorted, and containment weakened until a single triggering event exposed the instability.
The question we should have been asking all along
At some point, it became clear to me that we had been asking the wrong question. The real challenge was never how to make AI smarter. The real challenge was how to prevent intelligence from outrunning human authority.
That single shift changes everything. It reframes AI not as a tool that simply needs better answers, but as a force that must operate inside explicit boundaries, boundaries of authority, responsibility, evidence, and consequence.
In this framing, safety does not happen after execution. Governance does not sit outside the system. And accountability cannot be reconstructed after the fact. All of it must exist before action is allowed to occur.
The missing requirement no one is naming
Once you see the problem clearly, an uncomfortable conclusion follows. Any AI system that cannot make authority explicit is not governable at scale.
If a system cannot show who owns a decision before it acts, responsibility is already lost. If it cannot visibly distinguish between fact, inference, and projection, confidence becomes misleading by default. If credible opposition can be bypassed or silenced by speed, dissent becomes symbolic rather than protective.
And if harm prevention relies on policy documents rather than enforced limits, failure is only a matter of time. Authority is the first fracture point, but it is rarely the only one. When complex systems begin to drift toward failure, three things usually weaken together, clarity of authority, integrity of truth, and containment of consequences. When those three degrade simultaneously, instability begins to spread through the system long before failure becomes visible.
Why this still feels invisible
Many organisations believe they already address these issues. They point to ethics boards, governance frameworks, review committees, and post‑hoc audits. The problem is not intent. It is timing.
Most of these mechanisms live outside the moment of execution. They advise. They review. They explain. They do not enforce. By the time they engage, the action has usually already occurred.
This is the difference between preparedness and immunity. Preparedness assumes failure will happen and focuses on responding well. Immunity is designed so certain failures cannot propagate in the first place. As intelligence accelerates, that distinction becomes existential.
The cost of getting this wrong
When authority is unclear, harm does not announce itself loudly. It accumulates. It shows up as small decisions that no one feels fully responsible for. As defaults that no one remembers choosing. As outcomes that feel inevitable, even when they were not.
Over time, this erodes trust. Inside organisations. Between leaders and teams. Between companies and the people they serve. And when trust collapses, regulation follows. Not thoughtfully. Not precisely. But forcefully, in ways that often freeze innovation rather than guide it. If we want a future where AI is both powerful and trusted, we cannot leave authority implicit.
The future this makes possible
Imagine AI systems that are fast but bound. Systems where uncertainty is visible, authority is unavoidable, and decisions cannot outrun evidence. Not because humans behave perfectly, but because the system itself refuses to lie about what it knows, who is deciding, and what is at stake.
This future does not require fear. It does not require paralysis. And it does not require slowing innovation to a crawl. It requires a structural correction.
Solving this will require more than policy statements or ethics boards. It will require systems that make authority unavoidable, truth visible, and consequences containable before action is allowed to propagate.
The challenge ahead is learning how to recognise when complex systems are drifting toward instability, and intervening early enough to restore control. Until we make authority explicit, intelligence will continue to outrun us. Once we do, everything else changes.
Read more from Steve Butler
Steve Butler, CEO of Luminary Diagnostics
Steve Butler is the founder of the Execution Governance as a Service (EGaaS) category, architecting the future of intelligent, accountable enterprise. His work transforms risk from a reactive problem into a proactive, embedded safeguard against catastrophic failures like Drift, Collapse, and Pollution. As the Chief Strategy & Operations Architect, he proves that true autonomy can only be earned and must be governed by verifiable truth. He is also the author of multiple books that diagnose the fundamental illusions in the AI age and provide the solution: Sentinel, the Epistemic Citadel.
