Responsibility without visibility: why leadership exposure has quietly increased

By Dr Joanna Michalska

Most senior leaders are already accountable for decisions they did not personally make, cannot fully see end-to-end, and may struggle to explain when challenged.

That has always been partly true in complex organisations.

What has changed is how frequently AI now shapes decisions inside everyday operations, often without being experienced as a distinct system at all.

If something goes wrong today, could you explain how the decision was made?

Not the policy intent, not the ethical principles. The actual decision pathway that produced the outcome.

Accountability has not moved. Visibility has.

AI now influences recruitment, prioritisation, approvals, recommendations, and risk decisions through embedded tools, automated workflows, and third-party platforms. These systems rarely feel like “AI projects” once they are live. They disappear into business as usual.

Accountability does not disappear with them.

Boards, executives, CROs, and senior leaders remain responsible when decisions cause harm, generate bias, breach expectations, or fail under scrutiny. Yet their ability to see how those decisions are formed, how patterns evolve, and when intervention is required is often limited.

This creates a structural mismatch:

• Responsibility is explicit

• Authority is assumed

• Visibility is partial or delayed

  
  

That mismatch is becoming normal. It is also becoming dangerous.

When systems scale quietly, exposure scales with them

When AI-driven decisions scale, they rarely present themselves as a problem.

Often, there is no single failure point. Systems continue to operate. Outputs continue to be produced. From a technical perspective, everything appears to be working as intended.

What changes is not whether the system functions, but what it produces over time.

As decisions accumulate, patterns begin to form across hundreds or thousands of cases. Small biases compound. Edge cases become routine. Outcomes drift in ways that are difficult to detect when viewed individually, but material when seen in aggregate.

Because nothing appears broken, nothing is escalated.

The risk is not technical failure. It is the gradual emergence of behavioural and distributional effects that sit outside traditional monitoring and control mechanisms. These effects are probabilistic, cumulative, and often invisible to those closest to the system.

By the time they surface through complaints, audits, legal challenges, or public scrutiny, leaders are already in the position of explaining outcomes rather than governing them.

At that point, the question is rarely whether an organisation has principles, policies, or a framework.

The question is simpler and more exposing:

Could you see what was happening, and could you do something about it?

A simple starting point

If your organisation is deploying AI, start here.

Identify the operational workflows where AI is already influencing outcomes today.

Not pilots. Not future plans. What is live?

For each one, ask:

Who is affected by the outcome? 

What would unacceptable harm look like in this context? 

What would we need to see to detect early risk, not after damage has occurred? Who can intervene if needed, and under what conditions? 

What evidence would we rely on to justify a decision if it were challenged externally?

If you cannot answer those questions, you do not yet have governability.

You may have intent. You may have principles. But you do not have control.

Governance artefacts do not equal governance capability

Many organisations have done the “right” things on paper.

They have ethics statements. They have approval processes. They have training programmes.

None of these guarantees visibility into how systems behave once they are operating at scale. None of them ensures that someone can recognise when patterns are shifting in ways that matter. None of them ensures that authority exists to intervene quickly when required.

This is why accountability failures around AI often feel sudden when they become public. In reality, they have usually been building quietly over time.

Responsibility without visibility is not a philosophical concern. It is an operational condition.

This is not a technology problem

AI governance is often framed as a technical, ethical, or regulatory challenge.

In practice, it is a leadership and authority challenge.

Accountability and exposure already exist.

What is missing, in many organisations, is the ability to see how decisions are being shaped as they occur, to recognise when patterns begin to drift, and to act with authority before harm becomes structural, reputational, or irreversible.

This is not about better intentions or more detailed principles. It is about whether responsibility is matched by visibility and decision power in practice.

If something goes wrong today, could you explain how the decision was made?

If not, that is not a future risk you can plan for. It is the condition many organisations are already operating in.

Over the next 12 weeks, I’ll be exploring why this gap exists, why it is growing, and what leaders, boards, and executives need in place to govern AI systems with confidence rather than hope.

Join me in this conversation if you are responsible for decisions you cannot afford to explain after the fact.