TL;DR
AI honesty needs social permission
If saying "I used AI" makes good work look lazier, people will keep using AI and stop talking about it. That protects the individual and weakens the team.
Conor Donegan reports Atlassian Teamwork Lab research showing the same output can be judged worse when AI help is disclosed. Independent work summarized by Duke Fuqua points in the same direction: the fix is not better etiquette alone, but leaders visibly using AI, rewarding outcomes, and asking for review trails instead of confessions.
Attribution and thanks: this piece builds on Conor Donegan's Atlassian write-up of Teamwork Lab research, with corroborating context from Jessica Reif, Richard Larrick, and Jack Soll's AI social-evaluation research.
The work was identical
The cleanest part of the Atlassian finding is the controlled experiment. Nearly 1,000 U.S. knowledge workers evaluated the same written work. The variable was whether the hypothetical colleague disclosed AI help.
The penalty was large. The discloser was rated about 10x lazier and 24 percentage points less likely to be recommended for a high-visibility project. Treat the exact numbers as Atlassian's self-reported study data, but the direction is not isolated. Research by Jessica Reif, Richard Larrick, and Jack Soll found the same basic pattern: AI users can be judged less competent and less motivated.
That is the trap. If the work is good and the AI use is hidden, the person gets full credit. If the work is good and the AI use is disclosed, the person can lose status.
Better framing is not enough
Atlassian tested whether motivation changes the penalty. It does, a little.
Saying AI was used to support the team and client landed better than saying it saved personal time. Effort ratings moved from 45% to 56%. Recommendation likelihood moved from 43% to 51%. That is a meaningful lift. It still did not catch up to saying nothing about AI at all.
The lesson is not "find the perfect disclosure sentence." The lesson is that the status system is still miscalibrated.
Culture is the control surface
The useful part of Atlassian's piece is not the headline number. It is the moderator: culture.
In workplaces where AI use is visibly celebrated, the laziness penalty nearly disappears. AI users are rated as more efficient than peers who do not disclose.
The word visibly matters. A leader cannot say "use AI" in an all-hands and then evaluate work as if AI help is a stain. People take the real signal from promotions, praise, project selection, and what leaders admit doing themselves.
Underground AI is an operating risk
Hidden AI use protects the individual. It weakens the organization.
When AI work goes underground, teams lose the ability to learn from it. Good prompts do not become shared patterns. Useful workflows do not become team standards. Bad outputs are harder to catch because nobody knows where the tool touched the work.
That is the strategic risk. The company can win adoption and still lose the learning loop.
The standard has to move
The goal is not to celebrate every AI-assisted output. Some skepticism is earned. AI can hallucinate. It can sand off judgment. It can make weak work look polished.
The better standard is simple: AI use is normal; unreviewed AI use is not.
For leaders, the practical move is to model your own AI use in public. Praise good uses by name. Ask for the review trail, not the confession. Reward better outcomes, shorter cycles, and clearer judgment. Stop using visible effort as the moral proxy for work.
