The rapid deployment of AI across organisational workflows has created a governance gap. Technology capability has outpaced the development of ethical frameworks, accountability structures, and regulatory compliance mechanisms. Floridi et al. (2018) argue that the question facing organisations is no longer whether to use AI but how to ensure that AI use serves human interests, respects individual rights, and operates within boundaries that the organisation has consciously chosen.
The EU Artificial Intelligence Act (European Parliament, 2024) represents the most comprehensive regulatory framework for AI governance to date, establishing risk-based classifications and compliance requirements that affect any organisation operating in or serving European markets. For orchestrators, understanding the regulatory landscape is not optional. AI systems that were deployed informally may now fall under regulatory scrutiny, and the person directing the workflow carries a share of the accountability for the outcomes it produces.
What ethical governance means in practice
Ethical governance of AI is not an abstract philosophical exercise. At the operational level, it requires the orchestrator to address five concrete dimensions in every AI-augmented workflow they manage.
Accountability means that a named human being is responsible for the outcome of every AI-augmented decision. When an AI system recommends a course of action and the organisation follows that recommendation, someone must own the consequence. Daugherty and Wilson (2018) call these "responsible AI champions," individuals who ensure that AI systems are used in ways that align with the organisation's values and obligations. In practice, accountability requires that the orchestrator can explain how a decision was made, what role the AI played in it, and why the human chose to accept or override the AI's recommendation.
Transparency requires that stakeholders affected by AI-augmented decisions can understand, at an appropriate level, how those decisions were reached. This does not mean that every end user needs a technical explanation of how the model works. It means that people affected by AI-driven outcomes have access to information about what data was used, what criteria were applied, and what human oversight was in place.
Fairness means actively examining whether AI systems introduce or amplify bias in the decisions they support. AI systems trained on historical data will reproduce the patterns in that data, including patterns of discrimination. The orchestrator must ask whose perspectives are represented in the AI's training data, whose are missing, and whether the system's recommendations systematically advantage or disadvantage particular groups.
Privacy requires that AI-augmented workflows respect data protection requirements and the reasonable expectations of individuals whose data is being processed. The expanding capability of AI to infer personal characteristics from seemingly innocuous data points creates privacy risks that may not be immediately obvious. Zuboff (2019) describes the broader pattern of surveillance capitalism within which these risks sit, and the orchestrator must be attentive to the ways in which AI-augmented workflows may cross ethical boundaries even when they comply with the letter of data protection law.
Human dignity means that AI systems must be used in ways that respect the fundamental worth of the people they affect. Decisions that reduce individuals to data points, that remove the opportunity for human appeal or explanation, or that treat people as inputs to an optimisation process rather than as ends in themselves, fail this standard even when they are technically efficient.
Building an ethical reasoning practice
Ethical reasoning in the context of AI is a practice, not a single decision. Developing this practice means building the habit of asking a structured set of questions before, during, and after every AI-augmented workflow.
The Ethical Pre-Flight Checklist for AI-Augmented Decisions provides a practical framework:
Accountability: Is there a named human responsible for this decision? Can they explain how it was made and why?
Transparency: Could you explain to someone affected by this decision what role AI played and what human oversight was applied?
Fairness: Have you considered whether this AI system might systematically advantage or disadvantage particular groups? What data was it trained on?
Privacy: Does this workflow process personal data in ways that individuals would reasonably expect? Does it comply with relevant data protection requirements?
Human dignity: Does this use of AI respect the dignity of the people it affects? Would you be comfortable if the people affected could see how this decision was made?
Proportionality: Is the use of AI proportionate to the task? Are we using AI because it genuinely improves the decision, or because it is available?
The regulatory landscape
The EU AI Act classifies AI systems by risk level, from unacceptable risk (AI systems that are prohibited outright) to high risk (systems that require conformity assessments, documentation, and human oversight mechanisms) to limited and minimal risk. For many orchestrators, the most practically significant implications of the Act are the requirements for human oversight of high-risk AI systems, the obligation to maintain audit trails for AI-assisted decisions, and the prohibition on AI systems that manipulate users or exploit vulnerabilities.
The regulatory direction of travel is clear regardless of jurisdiction: AI governance is moving from voluntary best practice to legal requirement. Organisations that have built ethical governance into their AI workflows from the outset will find compliance straightforward. Those that have treated governance as an afterthought face significant retrospective work.
Ethical governance as a leadership responsibility
Strategic oversight of AI ethics cannot be delegated entirely to compliance functions. While legal, data governance, and compliance teams provide essential expertise, the orchestrator who manages AI-augmented workflows on a daily basis is the first line of ethical defence. They see how AI is actually being used, they observe the edge cases that policy documents do not anticipate, and they have the contextual understanding to judge when a technically compliant use of AI is nevertheless ethically questionable.
The most effective organisations build ethical governance into the fabric of daily operations rather than treating it as an overlay. This means integrating ethical review into workflow design, building ethical reasoning capability alongside technical AI skills, and creating the psychological safety for people to raise concerns about AI use without fear of being perceived as obstacles to progress.
Reflection prompts for practitioners
For each AI tool you use regularly, can you name the person who is accountable if it produces a harmful or biased outcome? If you cannot, what does that tell you about the governance maturity of your current AI practice?
Think about a recent AI-augmented decision in your area. If the people affected by that decision asked to see how it was made, including the role that AI played, would you be comfortable explaining?
When was the last time you raised an ethical question about the use of AI in your organisation? What made it easy or difficult to do so? What would need to change for ethical questioning to become routine?
References
Daugherty, P.R. and Wilson, H.J. (2018) Human + Machine: Reimagining Work in the Age of AI. Boston: Harvard Business Review Press.
European Parliament (2024) Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Brussels: Official Journal of the European Union.
Floridi, L. et al. (2018) 'AI4People: An Ethical Framework for a Good AI Society', Minds and Machines, 28(4), pp. 689-707.
Zuboff, S. (2019) The Age of Surveillance Capitalism. London: Profile Books.