We're using all of the AI's. We punish everyone who admits it.

In the spring of 2026, Atlasian's research department an experiment with 961 American knowledge workersSo let's say that this is the same thing as that. Everyone was given the exact same delivery — an e-mail that summarized a business proposal — from the same fictitious colleague. Everything was identical: text, quality, the sender.

One thing varied. A brief note on how the work was done. Some participants did not get any information about tools. Others were told that the colleague had used AI.

That's all it was going to be.

Participants stamped the colleague who admitted AI use, ten times more often as lat. They saw him as less princely. And they were 24 percentage points less willing to recommend him to a visible career building project. For identical work. The only difference was honesty.

The most honest thing is, it's called. Not in AI.

In June, Professor Morten Goodwin wrote in his father's friend that it is time to get rid of the KI barSo let's say that this is the same thing as that. I said that The shame is rational: It does not disappear because someone asks for transparency, but when systems stop punishing honesty. What I didn't know then was that two independent research environments had already weighed and measured the penalty — with the same result.

Conclusion first

The KI bar is not a position error you can appeal away from. It is a rational response to a real, documented and now numerical punishment. Three claims bear this text:

1. The scientists have measured the penalty for transparency, and it is brutal. The ‘late’ stamp is ten times more often. 24 percentage points fewer recommendations. And the find is not alone: peer-reviewed study from Duke researchers in PNAS The same punishment was found in four pre-registered experiments: those who assess the work of others, judge AI users as lazy, less competent and less demanding — and employ them less often.
2. We punish each other for what we all do. 94 percent of American knowledge workers spent AI last month. Nevertheless, those colleagues who admit the same thing are judging. We're not the victims of the KI-skam. We are the judges who maintain it.
3. The punishment is not inevitable. The same experiment took place where it disappears: in companies that actively celebrate AI usage, the latskas stairma collapses. And spin. The collegiate authorities shall consider the open as: more effective than those who remain silent. The shame is not a law of nature. It is a cultural product and leaders can build culture.

If you admit AI usage, you're stamped as lazy. For identical work.

The participants did not consider AI text against human-written text. All read the exact same text. The only thing that varied was if they were told that AI was used. However, when the judgment was more severe, there is only one explanation: that indication.

The researchers also tested whether the formulation helped. It helped. A little. The participants considered them as justification for the AI use in terms of the team and the customer, as more hardworking than those who said it saved them for the time being: 56 versus 45 percent of the effort, and eight percentage points higher in terms of recommendation. But both groups were still far behind those who didn't mention AI.

There you have the math of every piece of knowledge worker doing, knowingly or not: Right encroachment dampens the punishment. Tausity removes it. As long as it's true, silence is the rational choice.

This isn't one study. It's a pattern.

A supplier study, you can cancel. Atlassian sells collaboration tools and has an interest in the story. Therefore, it means that researchers at Duke University found the same punishment a year earlier and with a stricter method: four pre-registered experiments, 4,439 participants, published in the PNAS in May 2025. The findings form a chain:

1. People expects that others will judge them negative if they know they're using AI.
2. The expectation is justified. The person who gets help from AI is judged as being less qualified and less fulfilling than the one who gets the exact same help from others — or no help at all.
3. The punishment is not in your head. In an employment experiment, participants removed AI users.

Note the detail in the second paragraph: Help from a human being does not trigger judgment. There's something specific about AI help that does that. The Duke researchers' analysis points to the mechanism: Seen latskap is driving the sentence. We see AI usage, and we read character.

Two shades from the Duke study are included. The penalty was the strongest in those who did not use AI regularly. And it just dropped when AI was obviously the right tool for the problem. The penalty is not blind. It is a judgment of inappropriate tool-use, fields of people who don't see the tool as suitable yet.

You're not the victim. You're the judge.

Who's gonna hand this punishment out?

Not a remote control body. Not the IT department. The judges of both studies were common knowledge workers: colleagues like you and me. In Atlasian's parallel survey, 94 percent of American knowledge workers reported they spent AI on work last month. Two thirds do it regularly.

Read the two found together. Almost everyone uses AI. And in the same working life, the colleagues label the one who admits it, as lat. Atlassian does not report whether the judges in the experiment themselves were users. But then the punishment cannot come from the minority alone. Many of us are judging others for what we did in the previous job session.

Goodwin compared the KI talks to the talks about Se and Hör: Almost nobody admits to buying the magazine, but it is among the country's most sold. The analogy is good. But it's missing one term, and that term is the whole problem. We don't just keep talking about our own consumption. We're running on the nose of others. It's not just shame. It's a shame about the court authority.

Don't you recognize? Try yourself. Have you ever received a document, an AI in the formulations, and thought a little less about the sender — while your own AI assistant was standing open in the neighbouring tab? Then you've given the penalty. I've done it myself.

And I've got the sentence myself. In a recent INEVO leadership meeting I shared a bunch of AI-produced documents I had worked forward over time. The first question was not what I meant, but how long I had spent — and whether I had read the documents at all. The objection was partly deserved: I had shared too many documents, and volume without curing matter just prejudice. But notice the shape. The doubt went right on my efforts, not on content. It's research found in real-time, in its own team, in a company with good sharing culture and high AI levels. If the sentence falls there, it falls everywhere.

Duke shade gives hope: If the penalty is the strongest in those who don't use AI themselves, the sentence should be mild as the user share grows. But Atlassian conducted the experiment in the spring of 2026, in a workforce where almost all are users. The punishment was still there, in full strength. The water has spread faster than acceptable.

The option has been won. The acceptance has so far begun.

The spiral of shame

Put the pieces together, and you see a machine that drives itself. I call it the spiral of shame, and it has five steps that feed each other.

The Spiral explains the Goodwin observed: that the use of “shrink” in conversations, to “small language wash” and “a sparring partner for the text”. It's not a lie in the ordinary sense. It's rational reputation in a culture where full truth has a price no one will pay first.

And notice what the spiral does with Goodwin's solution. He asks us to be open. But the appeal is directed at that term in the spiral that has the least to earn from going first. Whoever breaks the silence alone pays full punishment, while the gain of transparency falls to everyone else. It's a classic collective action trap. No one can appeal out of it to an individual, just as little as anyone can appeal out of a queue to an individual driver.

For the organisation, the bill is larger than it looks. Where AI use goes underground, the company loses three things at the same time:

• ability to learn from what works
• ability to scale what is working
• ability to capture what goes wrong

Atlassian research leader Molly Sands precisely formulated the diagnosis: companies that ask employees to use AI while the employees punish each other to admit it, don't have an AI strategy. They have a contradiction.

Two reservations, because honesty is cheaper than upright

The first is the method. Both studies are based on assessments of hypothetical colleagues in design situations. Nobody measured what happens to real people in real coworkers' conversations over time. The selections are American, and Atlassian is a commercial player with an interest in the theme. I am also: INEVO lives, among other things, by helping companies with AI, so read me with the same scepticism. What makes the foundation solid, however, is convergence: two environments, different methods, one year between them — the same finding. And representative numbers confirm the direction: I Melbourne University Global Survey In 47 countries, 48 000 respondents admit 57 percent that they hide their AI use. People act like there's a penalty. There are rare people who are wrong about what is dangerous to admit at their own workplace.

The second reservation is more unpleasant: the sentence ‘letters’ has not been taken out of the air. Stanford researchers have documented workslop phenomenon: AI generated work that looks finished but does not pass the task on. Four out of ten office workers had received such a month and spent an average of almost two hours cleaning up each case. So the judges in the experiments have experience that gives prejudice to nutrition: Many provide AI-aided work spurvetSo let's say that this is the same thing as that. The punishment is unfair to the conscientious user, but it is not irrational as a group. The impact of the solution is clear: Transparency alone does not hold. Quality that someone is in for must be in charge. Otherwise, each admission confirms the prejudice it meets.

What breaks the spiral

The punishment is not a law of nature. It's the material's most important finding, and it comes from the same experiment.

The Atlassian researchers asked participants to describe their own workplace culture around AI, on a scale from the 'prohibited' to the 'feir'. So they compared the punishment across. In cultures that actively celebrate AI use — where managers and colleagues use the tools visually and raise the gains — the latskasstimma almost completely disappeared. And then it turned: where the participants considered the open colleague as more effective than the one that has been silent.

Read it again. Same confession. In one culture, it costs you reputation. In another one it strengthens it.

The punishment does not live in the AI usage. It lives in the culture that judges it.

And culture is something leaders can build.

A reservation is included here too: Culture described the participants themselves, so the found is a strong link, not a cause. The direction is however unique — and it points the same way as the Duke found.

What does it build? The material points to three things, in increasing order of effect.

The language is helping a little bit. To frame AI usage as something that serves the team and the customer, the punishment was mitigating. Start in own meetings: Review AI usage as method, not as confession.

Visibility helps more. The penalty disappeared where everyone uses the tools open. Each leader: shows its own AI workflow — including the failed attempts — the norm moves one square. The main sentence a manager can say in a professional meeting is not ‘they should use AI more’. That's "here I used AI for yesterday".

The systems decide. The shame is that we judge work from a guess about the process behind. The system solution is to stop guessing. In my teaching, I have for years evaluated students on observable signals from the work process — the drafts, questions, audits — instead of the final product being judged alone. Then students can use AI completely open, because transparency doesn't threaten anything: their thinking is already visible.

The same principle applies in working life, and it needs a rule people can be honest with. Difference tasks where the text only transports information — records, summaries, status reports, and tasks where a person commits to the content. Transport operations can take, open and out of date, without declaration and without raised eyebrows. The duties of commitment require man to think. There is the question "how did this happen?" at home, regardless of the tools.

See what the three have in common? None of them ask the individual coworkers to be braver. Everyone changes the cost of honesty. That's the difference between the appeal and the system.

Build the systems

Goodwin is right in that the KI-skamme is harmful, and he is right in that transparency is the goal. He also attacks a system when he criticises the government's KI ban in primary school. But to us adults he stops at the appeal: Be open. That's the order of the error.

Transparency is not the instrument. Transparency is the result.

Transparency arises when the leader has removed the penalty for honesty. Not before.

Now we know that the punishment is real, measured and brutal. We know that we're dividing it ourselves. And we know — from the same data — that it disappears and turns into cultures built for it.

The shame disappears on the day honesty lasts longer. The punishment is now an experimental finding. The cure is a very strong link. Both point the same way.

Build the systems.

PS: Of course I used AI for this article

You just read 2,500 words that people hide their AI use because honesty is being punished. It would be a little bit special if I ended up pretending that I wrote this alone with the spring pen.

So here's the admission: AI has transported this text from my head to your screen. The thinking, the positions, the source choices and the embarrassing meeting history are mine. The wordings have been in conflict with us together. The theme is too important for me to spend the days moving sentences around by hand. I my framework This is called a transport text, and transport texts must AI take. It's open. The responsibility for every word is nevertheless mine, and that is the part no machine can take over.

And now that you know it, you know what just happened to your assessment of me. There you have the whole article in one sense.

Sources

• Atlassian (2026): New research shows honey about AI use to work is backfireSo let's say that this is the same thing as that. Experiment with 961 American knowledge workers and heart rate measurement with 1 006 respondents.
• Reif, J. A., Larrick, R. P. & Soll, J. B. (2025): Evidence of a social evaluation penis for using AISo let's say that this is the same thing as that. PNAS, 122 (19). Four pre-registered experiments, four four hundred thirty-nine participants.
• Gillespie, N., Lockey, S., Ward, T., Macdad, A. & Hassed, G. (2025): Trust, attitude and use of artificial intelligence: A global study 2025So let's say that this is the same thing as that. University of Melbourne and KPMG. 48 000 respondents in 47 countries.
• Lower Hoffer, K., Rosen Kellerman, G., Lee, A., Liebscher, A., Rapuano, K. & Hancock, J. T. (2025): AI-Generated 'Worksloop' Is Destroying ProductivitySo let's say that this is the same thing as that. Harvard Business Review, in collaboration with Stanford Social Media Lab.
• Goodwin, M. (2026): Time to get rid of the KI barSo let's say that this is the same thing as that. Fatherhood Friends, 25. June.
• Lund, K. P. (2026): The KI bar is rationalSo let's say that this is the same thing as that. Fatherhood Friends, 3. July.