The AI Disclosure Problem at Work

You used AI to write that. You didn't tell anyone. Was that wrong?

I think most people who've felt that particular unease already know the answer, they just can't articulate why the guilt arrived when they broke no rule, produced good work, and delivered it on time. The wrongness feels real anyway, and that feeling is the whole story.

The Rules Haven't Caught Up

We're operating in a grey zone with no signage. Company policies range from Samsung threatening termination for using ChatGPT in 2023, to Shopify's CEO Tobi Lütke making AI use a formal performance expectation for every employee in April 2025, required before requesting more headcount. Goldman Sachs updated its conduct code to tighten AI rules. Apple restricted external tools over data concerns. Most companies have done neither, which means most employees are navigating this alone, quietly, without anything to reference.

A survey of more than 48,000 workers across 47 countries found that 57% of employees admit to concealing how they use AI at work. Not a small fringe. A majority. That number suggests a workplace culture caught between two pressures: the productivity gain is real and the tools are available, but the social meaning of using them is still unresolved.

Psychologists call the process people use to behave in ways that conflict with their self-image "moral disengagement," and normally it covers genuinely bad behavior. What's unusual here is that it's firing on behavior that isn't clearly bad, on work delivered on time, work the client or manager liked. People feel like they're getting away with something, even when no rule was broken. I think that's worth paying attention to.

Where Professional Identity Lives

I think the guilt isn't really about the AI. It's about where you've decided your professional worth lives.

For most of us, work identity is attached to effort. The IKEA effect names this: a 2011 study by researchers from Harvard, Yale, and Duke found people valued furniture they assembled themselves 63% more than identical pre-built pieces. Effort becomes part of the product's value, inseparable from it. The same logic runs through professional life. If I wrote that report, I'm competent. If the AI wrote it and I edited it, something feels borrowed, even if the output is better.

This is a genuinely strange position to be in, because the output is better, the client got better work, but the feeling is still there, because the effort that used to prove something to yourself isn't there anymore, and the proof matters more than you realized.

Research on disclosure makes this stranger. A study running more than 5,000 participants through scenarios where professionals disclosed AI use, in investment analyses, recommendation letters, corporate correspondence, found that explicit disclosure systematically reduced trust, not in the quality of the work, but in the person who submitted it. An HBR study found that when reviewers believed an engineer had used AI, they rated that engineer's competence 9% lower, despite reviewing identical work.

I find this genuinely difficult, because hiding AI use feels like a small deception and disclosing it invites an unjust penalty. We've built a cultural norm around effort as a proxy for skill, and now that effort can be partially offloaded, the proxy breaks down, and nobody has figured out what replaces it.

A Framework That Actually Works

I think most people are asking the wrong question. They ask "should I disclose that I used AI?" when the more useful question is: does knowing the method change how someone can use the output?

If a doctor uses AI to draft a diagnosis summary, another doctor reviewing it needs to know, because the interpretation and its limits depend on the method. If I use AI to draft a client proposal, the client probably doesn't need to know, because what they're evaluating is whether the proposal solves their problem. The quality of the recommendation doesn't depend on whether I typed it word by word.

A simpler version: disclose when the method changes how the output should be used or evaluated, because most of the time, it doesn't.

This isn't blanket permission to use AI invisibly. If your company has a policy, follow it. Courts are already requiring disclosure in legal filings, and academics operate under explicit rules. But in most professional work, what matters is whether the output is accurate, useful, and grounded in your judgment, and knowing the method doesn't change any of that.

The Question Under the Question

The guilt, I think, is a useful diagnostic, because it tells you where you've staked your identity. If the thought "what if they knew I used AI" triggers dread, that dread points somewhere specific, probably to the belief that your professional value lives in doing the hard parts yourself, even when the definition of "hard" is shifting under your feet.

I feel this too, in a specific way. There's work I've done with AI that felt lighter than it should, like a test I studied for and then was handed the answers, and the feeling doesn't prove anything about ethics, it's more like a stress test on assumptions I'd never examined: that effort equals skill, that time spent equals care, that doing it myself meant it was really mine.

Professions have always absorbed new tools. The spreadsheet arrived and did work that entire teams of accountants had done by hand, raising the same question about what the accountants' skill really was, and most of them kept their jobs, because the skill turned out to be judgment, not arithmetic.

Perhaps the guilt is ahead of its time. We seem to be watching professional identity negotiate a transition that hasn't finished yet, and the discomfort is what that negotiation feels like from the inside. The norms will catch up. Whether they catch up in ways that penalize people unfairly, or make room for what actually matters, probably depends on whether enough people are willing to have the conversation out loud.