Ethical AI in Practice: Where Principles Meet Operational Reality

Creating ethics statements is a straightforward task for many organizations. Many find that providing a reason why those closest to the work do not trust the proposed changes is much more challenging. During a recent Signal Room conversation, MarKeisha Snaith, a director of AI security governance and the founder of the strategy and advisory firm XIR, whose clients span government, healthcare, finance, and hospitality, addressed the gap directly. Culture is not what is posted on the walls, but is actually what is signaled by leadership over time: what gets rewarded, what gets escalated, what gets ignored, and what gets funded. An ethics statement can be generated in an afternoon, but the reality, or the exposure of the statement as a mere formality, will be created by the decisions of those several levels below the signers. At Hutchins Data Strategy Consultants, we see this most frequently. The governance language is present, but the actual behavior reflects a different reality.

Culture Is Built From Repeated Signals, Not Statements

Snaith argues every leadership decision or directive communicates some underlying message. As she says, when leaders prioritize speed over control, including at the expense of oversight, the culture becomes more risk tolerant. If leaders have a kill-the-messenger mentality when the source of a problem is made transparent, the organization's culture becomes defensive. If leaders incentivize cross-departmental collaboration, the organization's culture becomes collaborative and aligned. The culture of the organization is what employees observe happening most often over a sustained period; it is not what is displayed on the organization's website.

Snaith was direct and candid on the persistence of unhealthy organizational culture. When the primary leader is a poor leader, that poor leadership style becomes the standard across the organization, because the primary leader hires poor leaders. The organizational culture decays because the strongest employees depart for more positive employment opportunities, and the weakest employees remain because they are not employable by other organizations. The consequence of poor leadership and a poor organizational culture is that employees leave because the pay is poor and the leadership is poor.

This is why an infrequent town hall, or an even more sporadic ad hoc message, does not address the culture of the organization. As Chris Hutchins noted in the discourse, whatever action you take most often becomes the interpretation of trust by the people. Inconsistency will not reinforce a culture; it will drive the culture of the organization to an even worse state.

The Frontline Supervisor Is Where Ethics Becomes Real

One of the most useful, yet most overlooked, ideas in this discussion is also the most operationally relevant. The most significant danger in a transformation usually exists at the first level of supervision. This is the level that is closest to the front-line team that interacts with a patient, a customer, or anyone else on the other end of the work.

Snaith's logic is quite clear. If upper management fails to provide those supervisors with precise messages, clear guardrails, defined decision-making rights, and genuinely available support mechanisms, those supervisors will not be able to reinforce the organization's direction in a meaningful way. There is a possibility that they will actively oppose the behavior that upper management wants to see. There is also a possibility that they will encourage the behavior that upper management wants to see least. Seeing that there is an overwhelming tendency among individuals to trust the voices closest to them, a supervisor's uncertainty becomes the reality of the front-line worker, and lastly, the patient or customer.

This alters the traditional definition of ethical AI. It is no longer sufficient to state that AI will be developed with human oversight and fairness. There are now clear expectations that define, at the supervisory level, when a worker should intervene and override a system, when a worker should escalate, and where a worker should go to get a genuine answer. Hutchins identified this long-standing problem that has plagued the healthcare and technology industries: the only career path for a star technical worker is a management position for which that worker has received no preparation. The strategy does not fail at the boardroom. It fails at the front line, when an unprepared first-line supervisor is expected to carry it.

Training That Checks a Box Does Not Build Leaders

While Snaith is extremely blunt that most leadership development does not work, she does provide examples from her own career to clarify her opinion. While working at a financial institution, she was offered leadership training to accommodate a high performer who was at risk of leaving because they felt stuck. While well-meaning, the institution failed in the execution of this training. Over the course of a month, a designated internal HR person conducted training for approximately two hours each week, and it ended with a manager's sign-off that she had completed it. Snaith received very little from this. In her view, her promotion to the director level, along with her self-described progress as a natural leader and her external efforts, were largely responsible for her development, not the training the institution offered.

Snaith describes an ineffective training model that runs once a year. The program is aimed loosely at people who might become managers someday, with little to no follow-up after the training is conducted, mostly to check a box. This type of training has severe downstream impacts, as a large number of people in an organization become leaders who freeze on decisions, are unable to accept risk, and punish the people who are trying to break the mold and innovate. Snaith's recommendation is that training be conducted on a quarterly basis at minimum, and aimed not only at people who are managers but at the people in key decision-making roles now, starting with the senior executives whose behavior sets the tone for everyone below them.

Trust Is Generational, and the Margin for Error Is Thinner

Snaith attempted to describe these generational differences as a personal interpretation drawn from her research and coaching as opposed to empirical fact. Even with that disclaimer, her observations remain practical. Employees raised with continuous access to information expect a rationale for decisions, an opportunity to participate, and accountability at the senior levels. They disengage from what Snaith calls performative leadership. From what she says, Millennials expect opportunities for advancement along with the rationale for their assignments and continual feedback, as opposed to an unanticipated assessment months later. Gen X tends to prefer less talk and more results.

The common thread is that leadership builds trust by addressing the rationale for trade-offs and not framing things in the context of a need for obedience. She also posed a real question for the future: if younger employees disengage quickly and offer little forgiveness once alignment breaks, what does managing an enterprise look like when the room for error becomes this limited? The presence of the question is, in her opinion, more valuable than providing the answer.

Honest Messaging Beats Confident Messaging

Although both speakers had different experiences, they both described the same principle of leading through disruption. Executives often anticipate changes and understand how disruptive they will be, and there is tension between being completely open and having to time a message so that the business can continue to meet its obligations. But, as they both agreed, the safer failure is to say too little or to come across as evasive. People can accept a leader who states plainly the things they do not yet understand. People cannot follow a leader who they suspect is being untruthful.

Snaith also emphasized this point for the leader. For one, they should build a team of individuals who are smarter than the leader, but more importantly, it actually helps the team if you are open about what you do not know, because no one knows everything. She has observed leaders with too much ego for that, and it ends poorly for them because people inside of an organization communicate and notice things. This means more when combined with ethical AI. Responsible AI deployments claim to seek transparency, but that will not happen if the leaders cannot demonstrate it.

How Hutchins Approaches Ethical AI in Practice

Our work assumes that the ethics statement is the easy part and the operational reality is the job. We help healthcare organizations turn principles into the things that actually shape behavior: clear decision rights and guardrails at the supervisory level, escalation and override paths that frontline staff can use without guessing, and a cadence of leadership reinforcement consistent enough to set culture rather than decorate it. This is the same discipline that connects responsible AI to a strategy that survives contact with the front line, because a policy nobody is equipped to carry is indistinguishable from no policy at all — a point we develop in operational alignment. These themes run throughout The Signal Room podcast, where practitioners describe what culture, trust, and leadership take in the operational reality of healthcare transformation.

Authoritative sources

• WHO — Ethics & governance of AI for health
• NIST AI Risk Management Framework
• Agency for Healthcare Research and Quality
• NEJM AI
• U.S. Department of Health & Human Services — HIPAA