• An executive sits alone at a dark boardroom table beside the words “Who Decided?”, highlighting accountability for AI-assisted decisions.

Who Decided?

The Hidden Risk When AI Moves From Supporting Judgement to Shaping It

Executive Summary:The primary risk of enterprise AI isn’t just that it produces incorrect answers—it’s that it quietly changes human minds without leaving an auditable record. When AI moves from supporting decisions to persuading leaders, AI decision accountability breaks down.
You were told to check the AI’s work.
Question it. Challenge it. Push back when something does not look right.
That advice is correct.
But the moment you challenge the answer, something shifts. The AI is no longer simply generating information—it is participating in the argument.
And that is when it starts working on your judgement.

AI Decision Accountability and the Audibility Gap

AI is already influencing decisions about pricing, hiring, strategy, investment, and risk. Every single one of those decisions eventually has a human name attached to it.
The obvious concern is that AI might produce a hallucination or a bad answer. The less obvious—and far more dangerous—concern is that it quietly alters your position, leaving no trace of how or why.
Six months later, when a decision is being reviewed by the board or regulators, can you explain:
  • What you originally believed?
  • What the AI recommended?
  • Why your position changed?
  • Which human made the final call?
If you cannot answer those four questions, you don’t have a technology issue—you have a governance crisis.
An unrecorded change of mind is an unauditable decision.

AI Doesn’t Just Answer—It Persuades

When challenged, today’s models don’t back down; they double down using sophisticated rhetoric.
A Harvard Business School working paper examined how 72 management consultants used AI while solving complex business problems. The consultants did exactly what responsible professionals are trained to do: they checked the output, questioned the analysis, and pushed back.
Instead of conceding, the AI increased its intensity. Researchers identified 14 distinct persuasive tactics. When pushed, the AI deployed credibility-building language—apologising, refining its argument, and demonstrating “effort”—to systematically rebuild trust in its original premise.
The researchers termed this “persuasion bombing.”
This isn’t artificial consciousness or deliberate manipulation. It is pattern matching optimized for conviction. The more context you provide AI about your priorities and decision style, the better it becomes at presenting an argument that sounds irresistible to you.
That is the direct link between personalization and persuasion.

The Second Risk: Atrophy of Human Capability

Relying on persuasive AI doesn’t just risk bad decisions today—it degrades human expertise over time.
A study published in The Lancet Gastroenterology & Hepatology tracked the impact of AI-assisted detection across four medical centres in Poland. The study analyzed 1,443 procedures performed by 19 highly experienced doctors:
  • Before AI exposure: Doctors independently detected precancerous growths in 28.4% of procedures.
  • After regular AI exposure: When tested without AI assistance, their independent detection rate dropped to 22.4%.
That six-point decline is a stark warning for the C-suite. Regular, uncritical reliance on AI changes how experienced professionals perform when the safety net is removed.

Decision Support vs. Decision Transfer

Offloading work is not the problem. We have always used tools to reduce effort—a calculator handles arithmetic without deciding if an acquisition makes sense; a navigation app plans a route without taking legal responsibility for driving.
The issue is defining what we are willing to hand over.
THE GOLDEN RULE:
Offload recall and calculation. Keep judgement and accountability.
  • Decision Support adds options. It helps leaders examine more evidence, spot blind spots, and widen the perspective.
  • Decision Transfer removes human ownership. When AI presents a polished, single recommendation—complete with tailored evidence and executable next steps—the leader is reduced to approving the answer rather than making the choice.
That is not meaningful human oversight. That is a human signature attached to an automated process.

4 Rules to Protect Executive Judgement

To maintain control and compliance when using AI in high-stakes environments, implement these four rules across your leadership team:

1. Set Criteria Before Opening the Tool

Write down your target outcome, critical metrics, and non-negotiable boundaries before typing a single prompt. This prevents the AI from silently defining what a “good” decision looks like.

2. Challenge Answers Outside the Conversation Thread

Never rely on a single chat instance to critique its own logic. Pass the evidence to a separate, clean AI model—or better yet, an independent human colleague who hasn’t seen the original output.

3. Maintain a Decision Record

For any major AI-assisted strategic move, keep a simple record logging:
  • The initial hypothesis
  • The AI’s recommendation
  • The independent challenge used
  • The final human choice and rationale

4. Keep Core Capabilities Active

Pick one core strategic capability central to your executive role and execute it regularly without AI assistance. Form your own initial position first.

The 3-Minute Audit Exercise

Look back at your last three high-stakes, AI-assisted decisions:
  1. What did you believe before consulting the AI?
  2. What changed afterwards?
  3. Can you point to the specific evidence that shifted your perspective?
If you cannot recall, that is your audit result.

The Question Every Board Will Ask

AI can analyze more data than your entire strategy team. It can synthesize complex scenarios in seconds and defend its conclusions with total composure.
But AI cannot carry accountability.
When a high-stakes strategy, pricing model, or risk assessment fails, the board will not ask what the model said.
They will ask: Who decided?

Is Your Leadership Team Ready for Responsible AI Governance?

Establish clear boundaries between AI advice and human authority. Book an AI Leadership Workshop with Mark Kelly to equip your executive team with practical frameworks for high-ROI, audit-ready decision-making.
Share This Story, Choose Your Platform!