"AI ethics" sounds like a heavy, academic conference topic for someone else entirely — a slide deck full of abstractions that doesn't really touch your actual Tuesday. But the moment you start using these tools for real work, a handful of very ordinary, very concrete choices land on your desk without warning: whose data is this that I'm about to paste in, is this decision actually fair or just efficient, would this feel creepy to the person sitting on the other end of it if they knew? You don't need a philosophy degree or a company policy document to navigate these moments well. You need a small number of plain, memorable rules you can actually follow under time pressure, because that's when the shortcuts get tempting.
Consent: whose data are you feeding it?
Before you paste anything into an AI tool, pause for one second and ask who it actually belongs to. A customer's phone number and address, a colleague's private WhatsApp message, a client's confidential contract — none of these are yours to hand to a third-party service just because it's convenient and saves you five minutes of typing. The fact that a tool makes something easy doesn't make it yours to do.
The simple, reliable test: would the person whose data it is be comfortable knowing exactly what you did with it? If you're not sure, that uncertainty is itself the answer. When in doubt, strip out the personal details before you paste, use a private or offline model instead of a cloud service (see our earlier piece on running AI locally for exactly this situation), or simply don't do it at all. Convenience never legitimately outranks someone else's consent over their own information, however small the task feels in the moment.
This same principle applies even to data that feels entirely harmless on the surface. A colleague's draft performance review, a friend's half-finished business plan, a student's private exam answers — all of these carry a genuine expectation of privacy that doesn't simply evaporate just because AI makes it technically easy to process them somewhere else.
Bias: fair is not automatic
AI learns from the world exactly as it already is, biases fully included, because it has no other source of training material. Ask it to screen job candidates, rank people by "fit," or filter applications by some proxy for quality, and it can quietly reproduce and even amplify unfair historical patterns — while sounding completely confident and neutral the entire time. The smoothness of the output is precisely what makes this dangerous; a biased outcome delivered in calm, professional-sounding language doesn't announce itself as biased.
Keep a real, accountable human meaningfully in any decision that affects a person's life or livelihood, and sanity-check every output against fairness specifically, not just against efficiency or plain convenience. "The AI said so" is not, and will never be, an acceptable defence for a decision that ends up harming someone — the accountability stays with the human who acted on the output, exactly the same way it always has.
Five rules that keep you honest
Print these on a mental sticky note and actually refer back to it when a shortcut starts looking tempting:
- Don't feed in other people's private data without their clear, informed consent, however small the task feels.
- Don't present AI's work as entirely your own in situations where honesty and originality genuinely matter — an exam, a byline, a professional credential.
- Don't create fake images, audio or video of real people without their explicit permission, ever, regardless of how harmless the intent seems.
- Do keep a real human meaningfully in the loop for any decision that materially affects another person's life.
- Do disclose that AI was involved whenever staying quiet about it would genuinely mislead someone.
Trust is the real currency
Every single one of these rules exists to protect the same underlying thing: trust — the trust of your customers, your colleagues, your students, your audience, whoever is on the other end of whatever you're building. Trust takes years of consistent, ordinary behaviour to build, and it can be badly damaged by one careless, creepy misuse discovered after the fact. No productivity gain from cutting a corner is ever worth that trade, because the productivity gain is small and temporary while the trust damage is large and can last.
You don't have to become a philosopher or memorise a code of ethics to get this right in practice. Be the person who pauses for one extra second before pasting something sensitive, who keeps a real human meaningfully involved in decisions about other people, and who never fakes a person's likeness or voice without asking first. That's genuinely most of what responsible AI use looks like day to day — it's mostly just ordinary decency, consistently applied to a new and unfamiliar tool.
A worked example: the résumé-screening trap
Here's a concrete version of how the bias problem actually shows up, not as an abstraction. Imagine using AI to shortlist résumés for a role, and asking it to favour candidates who "fit the team culture." That instruction sounds neutral. In practice, "fit" is exactly the kind of soft, undefined criterion that tends to quietly encode whatever pattern already exists in the team it's trained to match — which can mean systematically disadvantaging candidates from different backgrounds, without a single explicit discriminatory word ever appearing anywhere in the process.
The fix isn't to avoid AI in hiring entirely — it's to replace vague instructions with specific, job-relevant criteria you'd be comfortable defending out loud to the candidates who didn't get through, and to have a human actually review the shortlist against those criteria before anyone's application is rejected. The same logic applies well beyond hiring: any time you're using AI to rank, filter, or judge real people, ask what specific, defensible criterion it's actually using underneath the surface, and whether you'd be genuinely comfortable explaining that exact criterion, out loud, to the person on the other end of the decision.
This same pattern shows up quietly in other contexts too — using AI to draft performance reviews, decide who gets a loan, or flag which customer complaints matter most. In every case, the fix is the same: name the specific criterion out loud, check it against fairness rather than just speed, and keep a human accountable for the final call.
We teach AI with these exact guardrails built into how we work every week, because powerful and responsible were never actually opposites, whatever the hype cycle suggests. Come see what that looks like in practice at a Sunday session — ₹99, and free if money is the only thing standing in the way.
Adapted and re-angled for the Institute of Applied AI from LearnPact's career blog. Authored under the LearnPact Faculty byline.
