Ethical smoothing AI governance: how moral disengagement and ethical fading allow AI governance concerns to be minimised, reframed, and silenced in not-for-profit, NGO, and community organisations in Australia
Ethical smoothing AI governance describes the process by which genuine concerns about AI adoption are progressively softened, reframed, or set aside in not-for-profit organisations, NGOs, and community organisations, until the concern no longer registers as a governance issue requiring action. Drawing on Bandura’s 1999 theory of moral disengagement identifying the cognitive mechanisms that allow people to act against their values without experiencing that contradiction, Tenbrunsel and Messick’s 2004 analysis of ethical fading showing how the ethical dimensions of decisions become invisible over time, and Umphress and Bingham’s 2011 research on unethical pro-organisational behaviour demonstrating how loyalty and mission alignment can drive ethically problematic conduct, ethical smoothing AI governance risk is directly relevant to NFPs and community organisations where mission commitment and resource pressure create conditions for concern minimisation. Ethical smoothing AI governance connects to psychological safety, stigma theory, normalisation of deviance, and security theatre in the NFH Lexicon, and underpins the NFH Behaviour Risk Card framework for not-for-profit and community organisations across Australia.
Ethical Smoothing
Ethical smoothing is not dishonesty. It is the gradual, often well-intentioned process by which concerns about AI get softened in the telling, reframed as manageable, and eventually absorbed into business as usual. By the time the concern disappears, nobody decided to ignore it. It was smoothed away.
Ethical smoothing AI governance: how moral disengagement and ethical fading allow AI governance concerns to be minimised, reframed, and silenced in not-for-profit, NGO, and community organisations in Australia
Definition
What is Ethical Smoothing in AI Governance?
Ethical smoothing describes the process by which genuine ethical concerns about AI adoption are progressively softened, reframed, or set aside as they move through an organisation. The concern begins as a real observation: an AI output was wrong, a client was affected, a process felt wrong. As it travels upward or outward, it gets qualified. Context is added. Competing priorities are noted. By the time it reaches the point where action would need to be taken, it has been transformed into something much less urgent, or it has disappeared entirely.
Ethical smoothing is distinct from deliberate suppression. It does not require anyone to consciously decide to ignore a concern. It operates through the accumulated effect of small reframings, each individually reasonable, that together neutralise the concern’s capacity to generate action. The people involved may genuinely believe they are being balanced and proportionate. The governance effect is the same as if they had chosen to ignore the concern from the start.
The Smoothing Process
Ethical smoothing typically follows a recognisable pattern. A concern is raised. It is acknowledged but contextualised: yes, but the tool is still in development. It is reframed: this is really a training issue, not an AI issue. It is deferred: we will review this in the next planning cycle. It is absorbed: all tools have limitations and we manage those limitations across the organisation. At each stage, the response is not unreasonable. Cumulatively, the concern has been smoothed out of the governance agenda.
Research Origin
Bandura (1999) identified moral disengagement as a set of cognitive mechanisms that allow people to act in ways that conflict with their own values without experiencing that conflict as a contradiction. His framework identified eight mechanisms, including moral justification, euphemistic labelling, and diffusion of responsibility, each of which operates to disconnect behaviour from its ethical implications. Applied to AI governance, Bandura’s mechanisms describe the specific cognitive moves through which concerns are reframed: the AI tool is positioned as serving the mission, problematic outputs are described in neutral technical language, and accountability is distributed across the team until it belongs to no one in particular.
Tenbrunsel and Messick (2004) developed the concept of ethical fading to describe how the ethical dimensions of decisions become progressively less visible over time and through process. Their research showed that framing a decision as a business decision, a technology decision, or an efficiency decision systematically reduces the salience of its ethical dimensions. The ethical content does not disappear from the decision; it fades from view. For AI governance, their analysis predicts that organisations that frame AI adoption primarily as an operational or productivity matter will find that ethical concerns raised early in the process have less and less traction as implementation proceeds.
Umphress and Bingham (2011) examined a specific and underrecognised driver of ethical smoothing: loyalty to the organisation and its mission. Their research on unethical pro-organisational behaviour demonstrated that employees who are highly committed to their organisation are more likely to set aside ethical concerns when raising them appears to threaten organisational goals. In mission-driven organisations, this dynamic is particularly potent. Staff who have dedicated themselves to the organisation’s purpose may smooth their own concerns about AI because naming them feels disloyal to the mission that AI adoption is framed as serving.
In AI Contexts
How Ethical Smoothing Shows Up With AI
AI adoption in community organisations frequently arrives with a productivity or efficiency rationale. The framing is operational. Staff who have concerns about client safety, data handling, or the quality of AI outputs find themselves raising ethical concerns in a conversation that has been structured as a practical one. The mismatch does not silence them directly. It repositions their concern as a different kind of objection, one that requires a different kind of response, and rarely the governance response the concern warranted.
Language as a Smoothing Mechanism
Euphemistic labelling is one of the most common smoothing mechanisms in AI governance contexts. AI errors become edge cases. Systematic biases become areas for ongoing improvement. Client harm becomes a suboptimal outcome. The language is not chosen to deceive. It reflects the vocabulary of technology adoption, where problems are framed as development issues rather than governance failures. The effect on accountability is the same regardless of intent.
Diffusion of responsibility compounds this. When an AI tool is vendor-provided, leadership-endorsed, and team-adopted, the ethical weight of any specific outcome is distributed across so many actors that it effectively rests with none. A staff member who observes a harmful output has no clear accountability structure to take the concern to. The smoothing that follows is partly structural, not just cultural.
The Mission Alignment Trap
Ethical smoothing is most powerful when AI adoption has been framed as serving the organisation’s mission. Concerns then carry an implicit counter-framing: raising them risks the mission-serving benefit the tool is delivering. Staff who are deeply committed to the organisation’s purpose are most vulnerable to this dynamic. Their own values become a mechanism for smoothing their own concerns.
Why It Matters for Not-For-Profits
The Specific Risk for Community Organisations
Not-for-profit and community organisations are structurally vulnerable to ethical smoothing for reasons that distinguish them from corporate or government contexts. Mission commitment is high. Resources are constrained. Leadership is often stretched across operational and strategic roles simultaneously. These conditions do not produce deliberate ethical negligence. They produce the exact pressures under which ethical smoothing flourishes.
When Mission Becomes a Shield
When AI tools are introduced as a way to do more with less, serve more people, or extend the reach of the organisation’s mission, concerns about those tools become harder to raise without appearing to oppose the mission itself. A peer worker who raises a concern about an AI-generated client communication is not opposing the mission. They are performing exactly the governance function the organisation needs. Ethical smoothing converts that governance contribution into an obstacle, and the conversion happens through the language of values rather than despite it.
Board and Leadership Vulnerability
Ethical smoothing also operates at board and leadership level. Boards of not-for-profit organisations are often composed of people with strong mission commitment and limited technology governance experience. When AI adoption is presented as an operational efficiency measure, the ethical dimensions of that adoption may not surface in board discussions at all. The framing determines what questions get asked. If the framing is operational, the ethical questions remain unasked, not because the board lacks integrity, but because the ethical content of the decision has faded from view before it arrived.
Ethical smoothing AI governance describes the process by which genuine concerns about AI adoption are progressively softened, reframed, or set aside in not-for-profit organisations, NGOs, and community organisations, until the concern no longer registers as a governance issue requiring action. Drawing on Bandura’s 1999 theory of moral disengagement identifying the cognitive mechanisms that allow people to act against their values without experiencing that contradiction, Tenbrunsel and Messick’s 2004 analysis of ethical fading showing how the ethical dimensions of decisions become invisible over time, and Umphress and Bingham’s 2011 research on unethical pro-organisational behaviour demonstrating how loyalty and mission alignment can drive ethically problematic conduct, ethical smoothing AI governance risk is directly relevant to NFPs and community organisations where mission commitment and resource pressure create conditions for concern minimisation. Ethical smoothing AI governance connects to psychological safety, stigma theory, normalisation of deviance, and security theatre in the NFH Lexicon, and underpins the NFH Behaviour Risk Card framework for not-for-profit and community organisations across Australia.
Observable Signals
What to Watch For
These patterns suggest ethical smoothing may be shaping how AI governance concerns are handled in your organisation. They are signals worth naming and discussing, not a definitive checklist.
- Concerns about AI outputs or AI-assisted decisions raised by staff are consistently acknowledged and then reframed as implementation or training issues rather than governance issues, with no record of the original concern, no clear owner, and no follow-up process that would allow the organisation to know whether the concern was resolved or simply absorbed.
- The language used to describe AI-related problems in team meetings, reports, or board papers is consistently technical and operational rather than ethical, with terms like edge case, calibration issue, or improvement area used in place of language that names the client impact, the accountability question, or the governance gap directly.
- Staff who raise concerns about AI tools are perceived, or perceive themselves, as resistant to the organisation’s direction rather than as performing a governance function, suggesting that the organisation’s culture has positioned concern-raising as opposition rather than as the essential error-correction mechanism that effective AI governance depends on.
Ethical smoothing AI governance: how moral disengagement and ethical fading allow AI governance concerns to be minimised, reframed, and silenced in not-for-profit, NGO, and community organisations in Australia
Ethical smoothing AI governance describes the process by which genuine concerns about AI adoption are progressively softened, reframed, or set aside in not-for-profit organisations, NGOs, and community organisations, until the concern no longer registers as a governance issue requiring action. Drawing on Bandura’s 1999 theory of moral disengagement identifying the cognitive mechanisms that allow people to act against their values without experiencing that contradiction, Tenbrunsel and Messick’s 2004 analysis of ethical fading showing how the ethical dimensions of decisions become invisible over time, and Umphress and Bingham’s 2011 research on unethical pro-organisational behaviour demonstrating how loyalty and mission alignment can drive ethically problematic conduct, ethical smoothing AI governance risk is directly relevant to NFPs and community organisations where mission commitment and resource pressure create conditions for concern minimisation. Ethical smoothing AI governance connects to psychological safety, stigma theory, normalisation of deviance, and security theatre in the NFH Lexicon, and underpins the NFH Behaviour Risk Card framework for not-for-profit and community organisations across Australia.
External References
- Bandura, A. (1999). Moral disengagement in the perpetration of inhumanities. Personality and Social Psychology Review, 3(3), 193-209.
- Tenbrunsel, A.E. and Messick, D.M. (2004). Ethical fading: The role of self-deception in unethical behavior. Social Justice Research, 17(2), 223-236.
- Umphress, E.E. and Bingham, J.B. (2011). When employees do bad things for good reasons. Organization Science, 22(3), 621-640.
Related Lexicon Entries
From theory to practice
Understanding the risk is the first step.
The NFH Toolkit gives your organisation practical tools to build governance processes that keep ethical concerns visible, create structured channels for concern-raising that do not depend on individual courage, and ensure that mission commitment strengthens rather than suppresses the organisation’s capacity to name and address AI governance failures. The Behaviour Risk Cards name this pattern in language your whole team can use.
Not For Humans: Because AI’s greatest risk isn’t technical, it’s Human.