Idris Faraday leads a nine-person operations team at a regional logistics company. One Tuesday his director dropped an idea in passing: "We've got AI now. Could it scan the team's Slack and email and flag who seems disengaged before they quit?" On paper it sounded efficient, even caring. Idris said he would look into it. That evening, sitting with it, the discomfort sharpened into a specific question: would he be comfortable telling his team to their faces that he was running an AI over their private messages to score their loyalty? The answer was no. That instinct, the unwillingness to say it out loud, turned out to be the most reliable ethics test he had. This lesson is about turning that instinct into something you can use deliberately.

What This Lesson Covers

AI is a tool, and like any tool some uses of it create real ethical concerns. This lesson is not a rulebook; rules cannot anticipate every situation. Instead it builds your judgment. We will name the recurring red flags so you can spot them quickly, walk a six-step framework for the genuinely gray cases, work a realistic recruiting example all the way through, and lay out how and when to escalate. The goal is that you leave able to make AI ethics calls in your own context with confidence.

A word on scope. This is team-level operational ethics: the calls you make about how your group uses AI day to day. Setting company-wide AI policy belongs to senior leaders. But "Legal and HR will handle ethics" is the comfortable lie that lets bad decisions through. Ethics is everyone's job, and a manager is usually the first and last line of defense.

Why This Sits On Your Desk

What is at stake is not abstract. Unethical AI use damages trust and reputation. It can harm real employees and customers. Ethical lapses carry long-term organizational consequences, and what you permit quietly becomes a statement of what you actually believe, regardless of what the values poster on the wall says. Idris understood that if he greenlit the disengagement scanner and it leaked, "the director asked for it" would not absolve him. He was the one who would have built it.

The most useful ethics test is simple: would you be willing to explain this decision, out loud, to the person it affects? If the honest answer is no, you already have your answer.

The Six Ethical Red Flags

Most problematic AI uses fall into one of six patterns. Learn to recognize them on sight.

• AI replacing human judgment in high-stakes decisions. "AI will decide who gets promoted, who gets the bonus, who gets fired." These calls affect livelihoods, demand values rather than pattern-matching, and using AI to make them quietly launders away your accountability. It crosses the line when AI makes the final call without human review, when the use is hidden, or when the affected person never knows AI was involved. Instead, let AI inform your thinking ("here is the data compiled") and explore options, but you make the call and you own it.
• AI on sensitive personal data without consent. Analyzing employees' private messages, scanning their social media, inferring personal characteristics. This violates privacy and autonomy, the inferences can be wrong and harmful, and laws like GDPR and CCPA may apply. This is exactly the line Idris's disengagement scanner crossed. Instead, be transparent about what you analyze and why, get genuine consent, limit use to the stated purpose, and protect the data.
• AI output that discriminates. A hiring or credit model that disadvantages certain groups, output leaning on stereotypes, training data that encodes historical bias. Discriminatory outcomes are harmful, often illegal, and hard to spot. It crosses the line when outcomes differ by protected characteristics and you know it but proceed anyway. Instead, test for bias, mitigate it, keep humans reviewing outcomes for fairness, document what you find, and if you cannot mitigate it, do not use AI for that purpose.
• Deception using AI. Generating convincing falsehoods, passing AI output off as human work where that matters, fabricating communications, impersonating people. It crosses the line when the deception harms someone, when disclosure would change how they respond, or when you knowingly spread false information. Instead, disclose AI assistance where it matters, verify facts before sharing, and do not hide AI use when honesty is at stake.
• AI for surveillance or control. Minute-by-minute productivity monitoring, tracking exactly when people work, auto-flagging "problem employees." This corrodes autonomy and trust, breeds a hostile environment, and usually backfires by making people less trustworthy, not more. It crosses the line when surveillance is secret, when data is used against people without due process, or when a power imbalance is exploited. Instead, focus on outcomes ("what did the team deliver?") rather than activity, be transparent about any tool you do use, and use it sparingly.
• Ignoring broader impact. Automating jobs with no plan for displaced people, brushing past environmental or downstream costs, benefiting yourself while others absorb the harm. It crosses the line when you see the negative impact and ignore it, or when the impact lands on people without a voice. Instead, ask who is affected, anticipate unintended consequences, and look for an approach that spreads benefit rather than concentrating it.

A Framework for the Gray Areas

Many real situations are not clean red flags. They are genuinely ambiguous, and those are the ones that need a method. Here is the six-step framework Idris used.

• Step 1, identify the concern. What feels ethically off? Is someone's autonomy affected? Could someone be harmed? Is there a fairness, honesty, or power-imbalance issue?
• Step 2, gather information. What do you actually know versus assume? What is the real impact, who is affected, what are the alternatives, and what happens if you proceed versus if you do not?
• Step 3, consult others. What perspectives are you missing? Talk to the affected people, to HR or legal, to colleagues from different backgrounds who may see what you cannot.
• Step 4, evaluate options. For each path, name the concern, ask whether you can mitigate it, and find the least harmful route.
• Step 5, decide and commit. Is this consistent with your values? Can you explain and defend it? Will it hold up over time?
• Step 6, implement with safeguards. Build in transparency, ongoing oversight, willingness to change course, and a clear owner accountable for the outcome.

A Worked Example: AI in Recruiting

Idris's company faced a sharper version of the same tension a quarter later. The opportunity: use AI to screen job applications. It could be faster and might even reduce some human bias, but it could also miss strong candidates or bake in new biases. Should they do it? He ran the framework.

Identify the concern: using AI in hiring decisions, with fairness and transparency at risk. Gather information: how much bias did the current manual process already carry, which candidates might be disadvantaged by a model, and what alternatives existed? Honest answer: the existing process was not bias-free either, which mattered. Consult others: he sat with the recruiting team on how it would change their work, ran the idea past a small candidate focus group, asked the diversity and inclusion lead whether it fit the company's equity goals, and checked compliance with legal.

Evaluate options. Option A, AI screens every candidate end to end, carried the highest bias and transparency risk. Option B, AI augments human review with a recruiter always making the final call, kept some AI-bias risk but contained it. Option C, AI only for certain roles or steps, was narrower but still introduced the tool. Option D, do not use AI, was safe but gave up the efficiency gain entirely.

Decide: Option B. AI surfaces and organizes information, a human recruiter makes every final decision, and outcomes are monitored for bias. Implement with safeguards: candidates are told AI is used in screening (transparency); candidate outcomes are reviewed monthly by demographic (oversight); if bias appears, screening pauses and is adjusted (willingness to change); and the recruiting manager owns the outcome by name (accountability). Notice that the framework did not produce a yes-or-no verdict on "AI in hiring." It produced a specific, defensible, monitored design. That is what good ethical judgment looks like in practice.

When and How to Escalate

You are not meant to carry every hard call alone. When you hit an ethical situation you are unsure about, escalate up a ladder. Start by discussing it with your team: "Does anyone else see an issue here?" If it is still murky, raise it with your manager or a peer leader. Beyond that, consult HR or legal on compliance and ethical considerations. If your organization has an ethics committee or values board, bring decisions that pit real values against each other to them. And when something carries genuinely organization-level implications, take it to executive leadership. Escalating is not a sign you cannot handle your job. It is part of doing the job well.

Anti-Patterns to Avoid

Five rationalizations show up again and again. Watch for them in your own thinking.

• Outsourcing ethics to AI. "The AI will handle the ethical complexity." It cannot. Ethics needs judgment. AI informs you; you decide.
• "Everyone does it." Peer behavior is not ethical guidance. Other companies' choices do not set your standard; your values do.
• Rationalizing. "We probably won't get caught," or "it's just a small impact." Small impacts accumulate, and the willingness to get caught is not the test. Would you defend it publicly is the test.
• Assuming consent where there is none. "It's in our privacy policy" is not consent. Real consent is informed, specific, and freely given, not buried in a document nobody reads.
• Claiming neutrality. "AI is objective, so our decision is objective." Choosing to use AI instead of human judgment is itself a value-laden decision, and the AI carries its own biases. Own the choice.

The Transparency Test

If you remember only one tool from this lesson, make it this. For any AI use on your team, ask whether you could explain it plainly to the customer it touches, and to a regulatory auditor. If you could not, or would not want to, that hesitation is data. It is the same instinct that stopped Idris on the disengagement scanner: the unwillingness to say it out loud. Honest AI use survives being explained. Use that as your everyday check, and escalate the cases that fail it.

Key Takeaways

• Ethics is your job, not someone else's. "Legal will handle it" lets bad decisions through. As a manager you are usually the first and last line of defense on how your team uses AI.
• Learn the six red flags. Replacing human judgment in high-stakes calls, analyzing personal data without consent, discriminatory output, deception, surveillance, and ignoring broader impact cover most problematic uses. Recognize them on sight.
• High-stakes decisions need a human in the seat. Use AI to inform and explore, never to make the final call on promotions, pay, or firing. Accountability cannot be delegated to a tool.
• Use the six-step framework for gray areas. Identify the concern, gather facts, consult others, evaluate options, decide, and implement with safeguards. It produces a defensible design, not just a yes or no.
• Transparency is the foundation. People have a right to know when AI affects them and how. If you cannot explain it openly, that is the signal something is wrong.
• Consent must be real. Informed, specific, freely given. Burial in a privacy policy is not consent.
• Escalate without hesitation. Team, then manager, then HR or legal, then ethics board, then executives. Asking for more perspective is part of responsible leadership.