New 2026 research: managers trust employees who use AI less, even when they asked for it. Why it happens, and how leaders and L&D can fix it.
Key takeaways
- In one reported experiment, employees who did the work themselves had about a 68% chance of earning their supervisor's trust. Employees who used AI had about 24%.
- The penalty showed up in behaviour, not just opinion. Leaders put real money behind their choices in a "rely or verify" trust game.
- The only thing that reliably shrank the penalty was an explicit organisational policy that encouraged AI use. When there was no policy, leaders punished AI users about as much as if AI had been banned.
- The penalty pushes people underground. A global University of Melbourne–KPMG study of more than 48,000 people in 47 countries found that 57% of employees hide their AI use and present AI-generated work as their own.
- This is not mainly an employee problem. It is a policy and management problem, and training can fix it.
Last updated · Bodhih Insights team · reviewed by Siddharthan VGJ, Founder
Their manager reads both reports and says nothing. But a week later, when a sensitive client brief needs an owner, the manager gives it to the other analyst.
Nobody would call that a decision, and the manager would probably deny it. Yet new research suggests it happens all the time. Managers who tell their teams to use AI often trust the people who do it less. Researchers call it a trust penalty, and it may be quietly undoing the AI training your organisation has paid for.
The Experiment That Should Worry Every AI Rollout
The paper is called Psychological Reactions to Subordinates' LLM Use at Work, by Samuel Meyers, Roman Briker and Yochanan Bigman. It is scheduled for The Leadership Quarterly (volume 37, 2026). Here is what makes it unusual.
Supervisors never judged the quality of the work. In the main experiments, they were simply told whether an AI tool had helped produce it. The quality was the same. Only the label changed, and the label was enough to lower trust.
Trust was measured with money. In a trust game, leaders chose whether to rely on an employee's work or spend effort verifying it, with real gains or losses at stake. They chose to verify AI users more often. So this goes beyond people saying they mind AI on a survey. Leaders acted on it.
Experience did not help. You might expect managers who use AI themselves to be more forgiving. According to the authors, supervisors' own AI experience made no consistent difference. Neither did how important the task was.
The effect was large. One of the authors reports an overall effect size of about 0.86 (Cohen's ) across the studies. In behavioural research, that is a strong effect.
A separate study in the same paper looked at real work: academic supervisors assessing graduate students' work. The pattern held. The more a student had relied on AI, the less their supervisor trusted them.
It Is Not Just Bosses: The "Lazy" Label
The Leadership Quarterly paper adds to an earlier finding. In 2025, researchers published four preregistered experiments with 4,439 participants in the Proceedings of the National Academy of Sciences (PNAS). They found that people who used AI at work were seen as lazier, less competent and less diligent than people who got the same help from a colleague or another source.
The more troubling part is that employees expected to be judged this way, and many chose not to use AI, or not to say so, because of it.
So there are two forces at work:
- Managers judge AI users more harshly, even when the work is just as good.
- Employees know it, so they use AI quietly or not at all.
Together, these produce a spiral that most AI training programmes never address.
The Silence Spiral
Here is how the trust penalty turns a well-meant AI rollout into a hidden risk.
| Step | What happens | What the organisation sees |
|---|---|---|
| 1 | Leadership announces an AI push and funds training | "We are an AI-first company" |
| 2 | Employees start using AI and mention it | Early wins, some excitement |
| 3 | Managers, without noticing, trust AI users less | Nothing visible: no one sees a bias in their own decisions |
| 4 | Employees notice who gets the important work | Quiet adjustment |
| 5 | AI use goes underground: used, but never mentioned | Usage data looks low; leaders think adoption has stalled |
| 6 | Hidden AI output goes unchecked | Errors appear with no clear source |
| 7 | Leaders conclude AI is risky and tighten the rules | The penalty gets stronger and the cycle repeats |
Step 6 is where the real cost appears. The same University of Melbourne–KPMG study found that 66% of employees rely on AI output without checking its accuracy, and 56% say they have made mistakes in their work because of AI. When AI use is hidden, nobody reviews it, nobody learns from it, and nobody can tell which mistakes came from the tool and which came from the person.
A 2026 study of consultants in the Journal of Intelligence put it plainly: fear of a competence penalty was linked to more hidden use of generative AI. The more people worry about being judged, the more they hide.
We have written before about Shadow AI, where employees use tools nobody approved, and about the unspoken AI rule, where managers expect AI use but never say so. The trust penalty connects the two. People hide AI because being seen to use it costs them something.
Why Managers Penalise AI, Even When They Asked for It
Most managers who do this are not against AI. Several older and very human habits are at work.
- 1. Effort used to be the signal. For most of working history, a polished report meant hours of work, and hours meant commitment. Managers learned to read effort as loyalty and capability. AI breaks the link between how a piece of work looks and how much effort went into it, but managers' instincts have not caught up.
- 2. "Who actually knows this?" When a manager learns that AI helped, a quiet doubt appears: if I ask this person a follow-up question, will they know the answer? This is a fair question, but it should be checked, not assumed.
- 3. Silence reads as prohibition. The study's most practical finding is that where organisations had no AI policy, leaders reacted as if AI were banned. "We never said you couldn't" does not protect employees. Unless the rule is written down, managers fall back on their old instincts.
- 4. Seniority bias gets in. A 2026 Microsoft Research experiment with 447 software engineers, in an organisation where AI use was already normal, found something encouraging and something not. AI disclosure no longer affected how engineers judged code quality or the author's competence. The author's seniority still did. Once AI becomes normal, the AI penalty can fade, but junior employees remain the most exposed. In many Indian organisations, juniors and freshers are the heaviest AI users.
What Fixes It: Judge the Work, Not the Tool
The research points to a clear set of fixes. None of them need new technology. They need clear rules and better management habits.
1. Write the rule down, and make it encouraging
This is the single strongest lever in the study. A policy that says AI use is allowed and expected, with clear limits, removed most of the penalty. A policy that only lists prohibitions, or no policy at all, leaves it in place. If your AI guidance is a two-line email from IT, it is not doing this job.
2. Make disclosure a normal part of the work
Create a simple, standard way to say how work was done, for example: "AI-assisted, human-verified: figures checked against source." When everyone uses the same wording, disclosure stops sounding like a confession. Research by Baum, Gerpott and Kerschreiter (2026) suggests the penalty shrinks when AI use is framed alongside tools people already accept without thinking, such as spellcheck or a calculator.
Reward verification, not just output
The worry managers really have is "Does this person understand the work?" So ask for the part that shows it: What did you check? What did you change? What did AI get wrong? An employee who can answer those questions has shown more judgement than one who typed every word by hand.
Train managers to review AI-assisted work
Most managers were trained to review work from people. Few have been trained to review work from people and AI. They need a short, practical checklist: where AI tends to make errors, which claims to spot-check and which follow-up questions reveal real understanding. This is a leadership skill, not a technical one.
Watch for the bias in allocation decisions
The penalty rarely shows in a performance review. It shows in who gets the next important assignment. Ask managers to notice whether the people they are choosing for visible, high-stakes work are the people who openly use AI, or the ones who don't mention it.
Protect juniors explicitly
Junior staff are the heaviest AI users and the most exposed to bias based on seniority. Mentors and team leaders should make clear that open, verified AI use is a sign of good judgement, not a shortcut.
Old Trust Signals and New Ones
| What managers used to read as "trustworthy" | What actually signals trust when AI is used |
|---|---|
| Visible hours and effort | Clear explanation of how the work was produced |
| Polished first drafts | Errors caught and corrected before submission |
| "I did it all myself" | "Here is what AI drafted and what I changed" |
| Never asking for help | Knowing when not to use AI |
| Knowing the answer instantly | Knowing how to verify the answer |
What This Means for HR and L&D Leaders
If your organisation has invested in AI training, the trust penalty may be cancelling part of it. Employees learn the tools, then hide them because using them openly costs them something. Here is a practical sequence.
- Audit your AI guidance against the research. Does it actively encourage use, or only list what is forbidden? Is it known by managers, not just employees? Is there a standard disclosure line?
- Train managers, not just staff. Most AI training is aimed at individual contributors. The research shows the bottleneck is the person evaluating the work. AI for leaders training and leading human–AI teams give managers the judgement to evaluate AI-assisted work fairly.
- Teach responsible use and verification together. Generative AI training and responsible AI training should cover how to show your checking, not just how to write prompts.
- Name the bias. Managers cannot correct a bias they don't know they have. Cognitive bias training that includes the effort heuristic and the AI penalty gives managers a vocabulary for something they currently do without noticing.
- Make it safe to talk about. Hidden AI use is a psychological safety issue. Psychological safety training and leading with trust help teams talk openly about how work really gets done.
- Rebuild how performance is judged. Update performance management conversations and feedback practices so they reward judgement, verification and outcomes rather than visible effort.
- Start before habits form. New managers set their review habits early. First-time manager training is the right place to teach "judge the work, not the tool".
Not sure where your teams stand? Our free AI readiness check takes a few minutes and needs no sign-up.
A Five-Question Check for Your Next Leadership Meeting
- If an employee told their manager today, "AI wrote the first draft," would that help or hurt them? Be honest.
- Does our AI policy say AI use is encouraged, or does it only say what is forbidden?
- Do our managers know how to review AI-assisted work, or are they relying on instinct?
- When we choose people for high-visibility work, do we know whether open AI users are getting a fair share of it?
- Do we have a standard, neutral way for people to say how a piece of work was produced?
The Bigger Story: Trust Has to Be Redesigned, Not Just Kept
For decades, managers built trust by watching effort. AI has broken that signal permanently. The organisations that get the most from AI will not be the ones with the best tools. They will be the ones whose managers learn to trust judgement, transparency and verification instead of visible effort.
That is good news, because those habits can be taught. The trust penalty is not a law of human nature. The research shows it shrinks when organisations write clear, encouraging rules, and the Microsoft study suggests it can fade once AI use is normal and openly discussed.
Until then, every organisation that pushes AI while leaving managers' instincts untouched is sending its people a mixed message: use AI, but don't let us see you doing it.
Bodhih designs practical AI-at-work and leadership programmes for organisations across India, in person and virtually, for both teams and the managers who lead them. Talk to us about AI training for your managers and teams.
Frequently asked questions
What is the AI trust penalty at work?
The AI trust penalty is the finding that managers tend to trust employees less when they learn the employee used an AI tool such as ChatGPT, even when the quality of the work is the same. A 2026 study in The Leadership Quarterly, involving more than 2,300 leaders across six studies, found this effect consistently.
Do managers really trust employees less for using ChatGPT?
According to the 2026 Leadership Quarterly study by Meyers, Briker and Bigman, yes. In one reported experiment, employees who did the work themselves had about a 68% chance of earning their supervisor's trust, compared with about 24% for employees who used AI. Leaders also staked real money on these judgements in a trust game.
How can organisations reduce the AI trust penalty?
The strongest factor in the research was an explicit organisational policy that encourages AI use. Other practical steps include a standard disclosure line such as "AI-assisted, human-verified", training managers to review AI-assisted work, rewarding verification and judgement rather than visible effort, and watching for bias in who gets important assignments.
Why do employees hide their AI use at work?
Many employees expect to be seen as less competent or less hard-working if they admit to using AI. A global University of Melbourne–KPMG study of more than 48,000 people in 47 countries found that 57% of employees hide their AI use and present AI-generated work as their own. Fear of a competence penalty is linked to more hidden use.
Is hidden AI use a risk for companies?
Yes. When AI use is hidden, its output is rarely checked. The same University of Melbourne–KPMG study found that 66% of employees rely on AI output without checking its accuracy and 56% have made mistakes because of AI. Hidden use also makes it impossible to learn where AI is helping and where it is causing problems.
Should employees disclose when they use AI?
In most organisations, yes, as long as there is a clear, encouraging policy and a standard way to disclose. Disclosure works best when it explains how the work was verified, for example "AI drafted the summary; I checked every figure against the source." Without a supportive policy, disclosure can carry a penalty, which is why leaders need to set the norm first.
What should AI training for managers include?
Beyond using AI tools themselves, managers need to learn how to review AI-assisted work, which claims to spot-check, which questions reveal genuine understanding, how to recognise their own bias against AI users, and how to set clear team norms for AI use and disclosure.
Where do these facts come from?
Meyers, Briker and Bigman, Psychological Reactions to Subordinates' LLM Use at Work, The Leadership Quarterly, vol. 37 (2026), as reported by the Hebrew University of Jerusalem and in press coverage in November 2026; Reif, Larrick and Soll, Evidence of a social evaluation penalty for using AI, PNAS (2025); University of Melbourne and KPMG, Trust, attitudes and use of artificial intelligence: A global study 2025; Microsoft Research (2026), study of AI disclosure in code review with 447 software engineers; Journal of Intelligence (2026), study of hidden generative AI use in consulting; and Baum, Gerpott and Kerschreiter (2026). Some figures come from press summaries of the research and should be read as indicative.
