The AI Trust Penalty: Why Managers Trust Employees Less When AI Is Involved, and How Leaders Can Make Honest AI Use Safe
The HCL Review PodcastNovember 18, 202600:25:31

The AI Trust Penalty: Why Managers Trust Employees Less When AI Is Involved, and How Leaders Can Make Honest AI Use Safe

Abstract: Organizations are spending heavily to put generative AI into employees' hands, yet a growing body of experimental research shows that using AI can quietly cost workers the trust of the people who evaluate them. A December 2026 study in The Leadership Quarterly found that supervisors accepted employees' self-reports 68% of the time when the work was done without AI but only 24% of the time when AI was involved, and that lower trust carried over into poorer ratings of performance, contribution, and promotability. Converging evidence from experiments on social evaluation, AI disclosure, and code review, together with large workforce surveys, shows that employees anticipate this penalty and respond by concealing their AI use. This article synthesizes that evidence, examines the consequences for performance, learning, fairness, and risk, and sets out five evidence-based organizational responses: making AI use explicitly legitimate, replacing confession-style disclosure with process transparency and verification, evaluating outcomes rather than effort signals, building manager fluency, and auditing who bears the penalty. It closes with three pillars for building calibrated trust over the long term.

Powered by the WRKdefined Podcast Network. 

[00:00:00] Picture this, right? You've got two analysts on your exec team. It's like a Friday afternoon, clock is ticking down. And they both hand in a market summary. Right. The classic Friday scramble. Exactly. And both of these summaries are flawlessly accurate. They're clear. They are right on time. But one of the analysts just happens to mention, like totally in passing, that she used the company's approved AI assistant to draft the first version.

[00:00:26] And then she, you know, personally verified every single figure. But the other analyst hands his in, says absolutely nothing about how he did it. Total radio silence. Right. So fast forward to Monday morning and your manager has to decide which of these two to trust with this massive high stakes board briefing. You think the honest one who followed the rules is the obvious choice.

[00:00:50] I mean, logically, yes. You would definitely think so. But based on the data we're looking at today, in a majority of organizations right now, that analyst who is totally transparent is actually at a severe disadvantage. Okay. We have to unpack this because that is just, it's the ultimate corporate paradox, right? It really is.

[00:01:10] So welcome to today's deep dive. We are diving into a brand new, just highly relevant piece of research. It's a November 2026 article from the Human Capital Leadership Review by Dr. Jonathan H. Westover. Yeah. And it's called the AI trust penalty. The AI trust penalty. And we aren't just looking at this one article either. We're cross-referencing Dr. Westover's findings with this massive stack of global surveys. We've got data from KPMG, Slack, Microsoft, Gallup.

[00:01:38] Plus some really fascinating peer-reviewed behavioral studies. Exactly. Just to see how this is actually playing out. Because our mission today is to figure out why doing exactly what leadership is begging you to do, which is innovate with AI, is somehow quietly tanking your performance reviews behind closed doors. Yeah. It's a huge disconnect. It's wild.

[00:01:59] So whether you're a manager trying to evaluate your team or an analyst turning in reports or just, you know, someone trying to survive the modern workplace, this data is going to completely change how you submit your very next project. It really fundamentally reframes how we talk about artificial intelligence at work. Because, I mean, we've spent the last three years watching executives urge employees to experiment, right? Right. Like use the tools. Innovate. Companies are spending millions on enterprise licenses running these pervasive adoption campaigns.

[00:02:29] So on paper, everyone is supposedly on board. But on the ground floor... On the ground floor, managers are unconsciously treating the use of AI as a signal that an employee is just, well, cutting corners. And they're punishing them for it. I really want to get into the hard data here. Because this bias is real and measurable right now. And there's this startling study from December 2026 in the Leadership Quarterly by Myers and colleagues that just, it puts a hard number on this.

[00:02:59] It does. And that study captures the dynamics so perfectly. So the researchers put participants in the role of supervisors. Okay. And the supervisors had to decide whether to accept an employee's self-reported work at face value or pay a cost. Meaning, you know, spend their own valuable time and energy to verify the work themselves. Right. Which nobody wants to do if they don't have to. Exactly.

[00:03:23] So when the supervisors believed the work was done entirely from scratch, like zero AI involved, they accepted the employee's account 68% of the time. 68%. I mean, that's a pretty solid baseline of trust. It basically means you assume your team is doing the work properly. Right. But the moment, the exact moment AI was disclosed as part of the process, that baseline just shattered. Wow. Trust plummeted to 24%. Wait, from 68 all the way down to 24? Yeah. A massive drop.

[00:03:53] That is a steep cliff to fall off just for admitting you used a tool that, frankly, the company probably provided to you in the first place. It's incredible, right? If you're listening to this and you've recently submitted an AI-assisted project, there is a very real chance your manager scrutinized your work like three times as harshly as your colleague who just kept their mouth shut. And what's fascinating here is that this distrust isn't just contained to that specific spreadsheet or that single task. What do you mean?

[00:04:23] The researchers found this massive bleed over effect. That initial drop in trust actually carried over into lower ratings for the employee's overall performance. Oh, wow. So it taints everything. Everything. It lowered their perceived contributions to the organization. And crucially, they were viewed as significantly less suitable for promotion. That is insane. So the evaluator stops trusting the document and they just start distrusting the person. Exactly. It stops being about, I need to double check these numbers.

[00:04:53] And it morphs into, I don't trust your potential or your work ethic or your character. Which is terrifying when you look at the sheer scale of this. Like, there's a global survey from Microsoft and LinkedIn from early 2024 showing 75% of knowledge workers were already using AI. Three quarters of the workforce. Three quarters. Yet, as recently as August 2026, Calib Tracking showed only 26% of U.S. employees felt their company had a clear, defined AI integration plan. Yeah.

[00:05:21] So you've got this massive chunk of the global workforce operating in a total systemic gray area. Right. They're using tools that their companies haven't clearly defined the cultural rules for yet. It makes me think of using a GPS on a road trip. Oh, that's a good comparison. Right. It's like we want the passengers in the car to think we just have this miraculous innate internal compass. But admitting we used the GPS somehow makes us a quote-unquote worse driver in their eyes.

[00:05:48] Even if we avoided a massive traffic jam and got there twice as fast. You got the result, but you're judged on the method. But I do have to push back a little here. I mean, if companies like Microsoft and Meta are actively buying these enterprise licenses, if the C-suite is pushing this tech so hard from the top down, why are middle managers acting like gatekeepers to the 1990s? Well, to answer that, we really have to look at the psychological mechanism behind the penalty. Because the technology itself isn't actually the issue. It's not? No.

[00:06:18] The friction comes from the human evaluator's outdated metrics for what actually constitutes hard work. We rely heavily on this cognitive shortcut called the effort heuristic. The effort heuristic. Meaning like our brains are just wired to judge the value of an outcome by how much sweat went into producing it. That's exactly it. There's a 2025 study by Reif and colleagues that dove deep into this social evaluation penalty. Okay. What did they find? They found that evaluators unconsciously use visible effort.

[00:06:47] So the time you spend at your desk, the visible stress, the manual grinding of data. They use that as a proxy for your commitment and your professional integrity. Oh, wow. So for decades, if you stayed at the office until eight at night, that visible suffering signaled to your manager that you were dedicated. Precisely. But when AI sweeps in and drastically reduces that visible effort, the proxy just breaks down. Evaluators read that lack of struggle as a character flaw.

[00:07:15] So they start rating the employee as lazier, less competent, less independent. Here's where it gets really interesting, though. If the output is flawless, like those two analysts we talked about at the beginning, why are we still so addicted to the struggle? Like, are managers literally just grading us on how much we suffered to produce a spreadsheet? On an unconscious level, yes. Oh. Because without clear organizational norms telling them otherwise, managers just fall back on their gut instincts. Right. The default setting. Yeah.

[00:07:45] Shilke and Reiman actually identified this in this sprawling 2025 study. They did 13 different experiments. They call it the transparency dilemma. The transparency dilemma. Right. Basically, disclosing your AI use actively erodes trust because of perceived illegitimacy. Since the workplace norms around AI are still so unsettled, using it feels like you're breaking an unwritten rule. Even if a written rule says it's perfectly fine. Exactly.

[00:08:13] So by being honest about your workflow, you're basically turning yourself in for a crime that isn't even on the books yet. That's a great way to put it. And they also found this really critical vulnerability for employees. What's that? If you get outed as an AI user by a third party, so let's say a colleague mentions you used it or like an IT audit flags your workflow, that destroys trust significantly more than if you had just voluntarily self-disclosed it. But wait, that assumes people actually know what they're evaluating.

[00:08:43] Couldn't this just be that people inherently don't trust computers? Like they assume the AI is going to hallucinate or make a math error. So they just reject the work. Well, the human automation literature actually separates those two reactions. What you're describing is called algorithm aversion. Algorithm aversion. Okay. Yeah. Where we just don't trust the machine itself. But the trust penalty we are discussing today is entirely different. This isn't about judging the algorithm.

[00:09:10] It's about judging the human who chose to use it. Oh, I see. Yeah. The goal in organizational psychology is something called calibrated trust. That means trusting a tool exactly as much as its actual reliability warrants. No more, no less. Makes sense. But right now, organizations aren't calibrating their trust in the tool. They're just indiscriminately nuking their trust in the employee. And employees aren't stupid.

[00:09:36] I mean, if honesty gets you penalized and efficiency is misjudged as laziness, people are obviously going to adapt. Oh, absolutely. Which leads us directly into the fallout of this dynamic. And this is where the real world consequences get just disastrous. Because when you punish honesty, you guarantee concealment. The data supports that completely. KPMG ran a global study in 2025 across 47 countries. Wow. 47? Yeah. Massive.

[00:10:05] And they found that 57% of employees are hiding their AI use and they're presenting AI-generated work as entirely their own. Just claiming it. Straight up. And Slack's 2024 data mirrors this perfectly. 48% of desk workers said they'd be highly uncomfortable admitting to their manager that they used AI for common tasks. Wait, okay. But if the senior managers are the ones enforcing the penalty on their direct reports, we also

[00:10:30] know from Charter's 2026 data that managers are actually 1.6 times more likely to conceal their own AI use than their workers. That's the irony of it all. Isn't this just a catastrophic game of corporate chicken? Pretty much. You have an environment where managers are secretly using AI to handle their own overwhelming workloads, right? Right. And then they're turning around and quietly penalizing their direct reports for doing the exact same thing simply because nobody wants to be the one caught violating that effort heuristic.

[00:10:59] It feels exactly like the prohibition era. Oh, for sure. Like leadership officially bans or just ignores the, you know, the booze, which is AI in this case. But meanwhile, everyone is secretly running a speakeasy right at their desk. Yes. And the scariest part is because it's all underground. Nobody is checking if the product is actually safe to consume. That frames the risk of shadow AI perfectly. When tool use is pushed underground like that, organizational learning is just completely destroyed. Right.

[00:11:27] The most significant risk to a company isn't that an employee used AI to draft a document. The risk is that the output went completely unverified because the employee was too scared to admit they used a tool in the first place. And we have seen this blow up spectacularly in the real world. Oh, big time. Think about the Mata v. Avianca case in 2023. Lawyers in New York submitted a brief filled with fake legal precedents generated by ChatGPT. Right. And they were severely sanctioned by a federal judge. Yes.

[00:11:55] Or look at Deloitte, Australia in 2025. They had to refund the Australian government $440,000. Why? Because an official report contained a completely fabricated quote from a court judgment just generated by AI. See, in both of those cases, the disaster wasn't the AI hallucinating. The disaster was the human failure to verify the work. Exactly. If you have to pretend you didn't use an AI assistant, you can't exactly turn to a colleague

[00:12:22] and ask, hey, can you help me double check these citations? I think the tool might be hallucinating. You can't ask for help if it's a secret. Right. You are forced to just cross your fingers, submit the work, and just hope no one looks too closely. And beyond the legal and financial risks, there is a systemic fairness gap here that the sources highlight. This penalty does not hit everyone equally. No, it really doesn't.

[00:12:48] A 2025 study by Akar and colleagues examined over 1,000 software engineers who were reviewing identical Python code. Okay. Identical code. Right. And when the reviewers believed AI had been used to help write the code, the competence ratings for the engineer dropped by 9% overall. But the penalty hit women and older workers the hardest. Wait. So for the exact same lines of code, certain demographics are paying a higher professional

[00:13:15] price just for the perceived use of the tool? Yes. And the study dug into who was actually handing down these harsher judgments. Who was it? The harshest evaluators were senior male engineers who had not adopted AI themselves. Oh, wow. They were enforcing a penalty based on their own unfamiliarity with the tool, essentially shifting the goalposts on diverse talent that was actually just trying to innovate. So you have senior people who refuse to learn the new tech, punishing the talent that is

[00:13:45] actively trying to evolve. We can't stay in the Prohibition era forever, right? We need a way out. Thankfully, Dr. Westover's article outlines this evidence-based organizational antidote to dismantle the trust penalty. Yeah. We have to look at how companies can actually evolve their culture to fix this. So what's the first step? The foundation of fixing this is making AI use explicitly legitimate.

[00:14:11] Returning to that Myers study, they found that when leadership explicitly encouraged AI use, the trust penalty vanished. Completely vanished. It wasn't just reduced. It became statistically insignificant. Because silence from leadership leaves managers relying on their outdated effort heuristics. Saying nothing is functionally the same as banning it. I look at a company like Shopify. Back in 2025, their CEO, Toby Lutka, didn't just say AI was allowed.

[00:14:38] He issued a memo demanding reflexive AI use as a baseline expectation. Which is brilliant. Right. He made it clear that before a team asks for more headcount or resources, they have to prove they tried to solve the problem with AI first. That completely removed the ambiguity. It kills the stigma by making the tool a requirement rather than a shortcut. But, you know, once it is legitimate, an organization has to change how people talk about using it.

[00:15:05] We need to replace confession with process transparency. What does that look like in practice? The conversation has to shift from this guilty, I used AI, to a confident, here is how I verified the AI. Oh. I love that. That takes the spotlight off the tool entirely and puts it back on the human's judgment, which is what actually matters. Exactly. Judge Brantley Starr in Texas provided a great model for this. He issued a federal court order in 2023 that didn't ban AI in his courtroom.

[00:15:34] Instead, it required lawyers to formally certify that any AI-drafted language was checked by a human against traditional authoritative sources. Okay. So verifying is the mandate. Right. And we see similar models in healthcare. Kaiser Permanente rolled out AI medical scribes and made the process totally transparent to patients. Oh, that's interesting. The AI drafts the clinical notes, but the doctors are explicitly responsible for verifying and editing every single line.

[00:16:02] The transparency is just built into the workflow itself. Well, if you're verifying the work, that means companies have to evaluate outcomes, not effort. Yes. We have to stop judging people by the bags under their eyes. I saw this play out with Duolingo. They tried forcing AI metrics directly into performance reviews in 2025. Oh, that was messy. It was a disaster. They tied evaluations to how often people use the tools and even paused hiring contractors based on projected AI efficiency.

[00:16:28] It caused this tremendous internal and external backlash. They really had to walk that back. They did. They had to backpedal a year later, admitting that AI use shouldn't be forced where it just doesn't fit the actual task. You can't just mandate that people use a hammer when they need a screwdriver just to meet some arbitrary AI quota. Exactly. Conversely, you have companies like Meta and Microsoft successfully making AI impact a formal component of their review structures. And the key word there is impact.

[00:16:58] Right. Did it work? Did the tool help you achieve a better outcome? Did you exercise good judgment? You shouldn't earn high ratings just for logging into an AI portal 100 times a week. No, of course not. And you shouldn't be penalized for doing a task manually if manual execution yields a better result. Okay. So what does this all mean for the managers who actually have to sit down and do these evaluations? They need fluency.

[00:17:22] That Reef study we mentioned found that managers who are frequent AI users themselves judge other AI users much more fairly. Because they actually understand how it works. Because they understand the reality of the tool. Familiarity breeds calibrated trust. But I will say walking the tightrope between building fluency and becoming a tyrant is really difficult.

[00:17:44] How does a company walk that tightrope between building fluency, like say Zapier, and becoming a tyrant building fake compliance? Well, Zapier took a highly successful approach by running company-wide hackathons. They gave people hands-on practice in a safe, collaborative environment. And it worked, right? It did. They boosted their AI use from under 10% to over 50% in a matter of weeks. Wow. And then on the flip side of that tightrope, you have Coinbase in 2025.

[00:18:14] The CEO bought enterprise AI licenses and basically issued an ultimatum. Onboard by Friday, or you're fired. Yeah, that was extreme. They actually let engineers go who didn't adapt fast enough. I mean, coercion might build compliance, but it certainly doesn't build calibrated trust. No, and when managers are forced into compliance like that, they don't learn what the AI is bad at, which is a critical part of fluency. That's a great point.

[00:18:38] When managers truly understand a tool's limitations, they stop seeing it as a magic button for lazy people, and they start seeing it as an instrument that requires genuine human skill to wield correctly. Which means organizations have to continually audit the penalty. If you aren't looking at your internal data, you will miss the fairness gaps, like what happened with that Python coding study. You have to track it.

[00:19:00] Companies need to track who is using the approved tools, break it down by demographic, and actively investigate if certain groups are getting lower ratings after disclosing AI use. And when mistakes inevitably happen, the organization has to hold the line on accountability without reintroducing the stigma. Like the Butler-Snow law firm case. Exactly. That's a prime example. The lawyers were sanctioned by the court not because they utilized a generative tool, but because they failed to verify the citations it produced.

[00:19:29] Accountability was tied to the human failure of judgment, not the technological choice. So if we pull all of these threads together, what does the future of work actually look like for you, the listener, if a company actually gets this right? Well, if we connect this to the bigger picture, evaluating AI at work is ultimately about evolving what we value in human intelligence. For long-term calibrated trust, organizations really need three pillars. Okay, what are they?

[00:19:55] First, psychological safety, so people can admit to using AI and more importantly, admit when the AI makes a mistake without fear of retribution. Crucial. Second, AI-aware talent systems that rewrite competency models to reward critical evaluation, editing, and judgment rather than just visible effort and keystrokes. The effort heuristic is dead. Hopefully. And third, continuous governance loops, meaning constantly tracking for concealment and updating policies as the technology evolves.

[00:20:24] The workforce is moving away from a reality where sheer visible sweat is the primary measure of human value. And that brings up a really profound question for you to chew on as we wrap up today's deep dive. Dr. Westover's research mentions this idea of the AI ghostwriter effect. Yeah, that nagging feeling of ambivalence. Right. That feeling we get when a machine does the heavy lifting for us. Let's say we fix the trust penalty tomorrow. Let's say your company implements all these transparency labels. You diligently verify the work.

[00:20:53] Your manager praises your efficiency. Everything is completely above board. Best case scenario. What happens to your professional identity five or ten years down the line? If an algorithm drafted that flawless market summary and you just verified the figures, how do you define your own career legacy? That's a tough question. When you look back at your resume a decade from now, will you feel a deep sense of pride that you actually did the work? Or will you feel like you were just the editor-in-chief of an algorithm?

[00:21:20] It raises an essential question about where we derive our sense of professional self-worth when the physical struggle of the work is just removed. So the next time you're sitting at your desk on a Friday afternoon, staring at a flawless project, ask yourself, are you being penalized for your honesty? And more importantly, whose work is it really? Thank you for joining us to unpack the sources today. Keep questioning the world around you and we will catch you next time. Keep questioning the world around you.