There is a phrase I am hearing more often in leadership conversations, and it deserves a closer look.

"The system flagged it."

"The model recommended it."

"The agent handled it."

Each phrase is technically accurate. Each one is also, quietly, a small act of displacement. Not deception; most leaders saying these things are genuinely thoughtful people working in genuinely complex environments. But something is shifting in the architecture of how decisions are made, and with it, something is shifting in how accountability is held.

As AI becomes more embedded in how organisations think, decide and act, accountability is not disappearing. It is becoming harder to locate.

‍

Most of the conversation about AI in organisations is operational. Which tools. Which use cases. Which workflows. Which risks. These are necessary conversations, but they tend to obscure a quieter pattern that I think deserves more attention.

Across the organisations I work with, I am noticing a subtle redistribution of ownership. Decisions that used to sit clearly with a person are now diffused across a sequence of human and machine inputs. A recommendation comes from a model. A workflow is executed by an agent. A judgement is made by a team that has, in part, been shaped by what the system surfaced. By the time an outcome lands, the question of who chose it becomes genuinely difficult to answer.

This is not a failure of AI. It is a feature of how AI integrates with existing organisational dynamics. And it is happening fastest in cultures that were already, before AI arrived, slightly uncomfortable with ownership.

That is the pattern worth naming. AI is not creating the avoidance. It is offering it new and more sophisticated forms of expression.

‍

There is a long-standing assumption in management thinking that accountability is something you can engineer. RACI charts. Clear lines of reporting. Defined ownership. These tools have their place, but they have always rested on a quiet truth that we tend to under-articulate: accountability is not assigned. It is chosen.

You can give someone a role. You cannot give them ownership. Ownership is a relational act. It is something a person does in relation to other people, to the consequences of their decisions, and to the system they are part of. It requires the willingness to stand somewhere and to say, "this was mine," or "this is mine to hold." It requires what I think of as adult-to-adult ways of working, where responsibility is not handed down through hierarchy but taken up through relationship.

This is why accountability cannot be transferred to a model, an agent or a system. AI can make decisions based on the data and the framework it has been given. It cannot hold relational consequence. It cannot navigate the courage required to own a difficult call. It cannot read the power dynamics in a room and choose to stand inside them rather than around them. Those capacities live in humans, in the space between people, and in the system dynamics they collectively shape.

When we treat accountability as a mechanism that can be distributed across human and machine actors, we are not solving a design problem. We are quietly removing one of the load-bearing walls of organisational trust.

‍

This matters because trust is the infrastructure on which collective capacity is built. Not trust as a warm feeling, but trust as a working assumption that the people around me will hold what they have said they will hold, and that I will do the same.

When accountability becomes diffuse, trust does not collapse dramatically. It thins, slowly. The team meeting where no one quite owns the recommendation. The decision that everyone subtly implies was made by the data. The escalation that loops without landing. None of these moments are catastrophic in isolation. Cumulatively, they erode the sense that the system is held by people who are willing to stand in it.

AI accelerates this thinning when it is dropped into cultures that were already operating from compliance or pure performance logic. If we look at this through the lens of Spiral Dynamics, organisations operating predominantly from a Blue(compliance, rules) or Orange (performance, KPIs) worldview tend to treat accountability as something the system enforces. Add AI to that, and accountability becomes something the system also absorbs. Decisions become outputs. Outputs become metrics. Metrics become the answer to the question"who decided this?"

Barrett's values framework offers a complementary lens. At the higher levels; the ones associated with purpose, integrity and service, accountability is not extracted through pressure. It is chosen, freely, because the people in the system understand themselves as stewards of something that matters. AI in those cultures becomes a tool for clarity. AI in cultures still operating below that threshold tends to become a tool for diffusion.

This is the deeper observation. AI amplifies the design of the system it enters. In a culture of chosen accountability, AI sharpens it. In a culture of avoidance, AI makes the avoidance more efficient.

‍

Most of the public conversation about AI and accountability is framed as a question about machines. Can AI be accountable? Should AI be accountable? How do we govern AI?

These are important questions. But I think they are not the most urgent ones for leaders inside organisations. The more urgent question is this: are we designing systems where humans can still consciously choose accountability, or are we quietly engineering elegant ways to abdicate it?

This shifts the conversation from one about technology to one about culture, leadership and the relational fabric of the system. It asks us to look at AI not as something that happens to our organisation, but as something that reveals the organisation we already have.

It also asks something of leaders that is harder than it sounds. Not to govern AI more tightly, but to hold accountability more visibly. To resist the seduction of diffusion. To name, in the moments when it would be easier to say, "the model recommended it," that a human chose to act on that recommendation,a nd that the choice carries weight.

‍

I am wary of turning this into a checklist, because I think the move that is needed here is not procedural, but relational and dispositional.

Stewardship leadership, as I understand it, is the practice of holding the system on behalf of those who depend on it, including the people inside it, the people it serves, and the people who will inherit it. In an AI-enabled organisation, stewardship asks leaders to be unusually clear about where accountability lives, even when the architecture of decision-making is becoming more distributed. It asks leaders to design the spaces where chosen accountability is still possible; meetings, conversations, reviews, escalations, etc. and to model what it looks like to stand inside a decision rather than around it.

It also asks leaders to take seriously the relational work that AI cannot do. The honest conversation about what we are each ready to let go of, and what we are not. The dialogue about where control feels comfortable and where ownership feels exposing. The recognition that, as Maslow and Barrett both suggest in different languages, much of what we call "management" is actually a quiet negotiation with our own need for control, esteem and significance.

AI does not change this. It surfaces it.

‍

There is another layer to this that I am still sitting with.

If accountability becomes harder to locate in AI-enabled systems, it may not only be because decisions are more distributed. It may also be shaped by the conditions under which those decisions are formed.

The quality of what we decide is influenced long before the moment of choice in how we think together, whose perspectives are present, and how much space is given to exploring a question before moving to resolution.

As AI increases both the volume of information and the speed at which we can act, I find myself wondering whether our capacity for collective thinking is developing at the same pace. In some organisations this may already be visible, in others it may still be emerging. Without deliberate attention to this, speed can begin to outpace coherence and when that happens, accountability has less ground to stand on.

‍

- As automation increases inside your organisation, where is accountability becoming clearer and where might it be quietly dissolving?

- And what would it take, in your own leadership practice, to meet accountability as a chosen act rather than a diffused one?

I would be genuinely interested to hear whatothers are noticing. The patterns are not yet fully formed, and the conversation feels more useful when it is held collectively.

For now, I think the most useful thing we can do as a community of leaders is keep naming what we see clearly, and without rushing to resolve it.

‍

I drafted this piece in collaboration with Claude, working from my own notes, frameworks and months of observations. The thinking is mine; the structural shaping is shared. I draw on Spiral Dynamics (Beck and Cowan) and the Barrett Values Framework as ongoing reference points in this work. I mention both because writing publicly about AI integration whilst hiding my own engagement with it, or working from established frameworks without naming them would not sit easily with the chosen accountability this article asks for.

‍