Research shows that employees face a trust penalty when supervisors discover they use artificial intelligence, often leading to lower performance evaluations and missed promotions. Because many organizations fail to provide clear guidance, workers frequently conceal their artificial intelligence use, which severely damages workplace transparency and hides important operational errors. To fix this cycle, leaders must officially legitimize tool adoption, evaluate staff based on actual results rather than visible effort, and build managerial fluency to ensure fair treatment. Furthermore, organizations need to monitor demographic disparities to prevent these hidden biases from disproportionately harming vulnerable workers. Ultimately, establishing calibrated trust and process transparency allows companies to safely integrate new technologies while maintaining professional accountability.
Powered by the WRKdefined Podcast Network.
[00:00:00] Welcome to the Debate. If you look at the history of human craftsmanship, there is this, I don't know, almost sacred relationship between the time spent on a task and the value earned from it. Right, the whole blood, sweat, and tears thing. Exactly. I mean, you carve a chair by hand over three weeks, and it is considered a masterpiece, right? But you build the exact same chair in a factory in three hours, and suddenly, well, there's just furniture. Yeah, because we are deeply conditioned to equate struggle with virtue. We want to see the sweat on the brow.
[00:00:29] It reassures us that the person actually cares about what they are delivering. It totally does. But then, and this is where it gets interesting, you step into the modern knowledge economy, and we are handing everyone the ultimate cognitive power tool, generative AI. And suddenly, that old equation is completely short circuiting. Exactly. Organizations today are investing massive amounts of capital, right? Just dumping money to put these tools in the hands of their employees.
[00:00:52] Yet, there is this December 2026 study published in the Leadership Quarterly, and it revealed a staggering phenomenon. Researchers are calling it the AI trust penalty. Oh, the trust penalty is fascinating and brutal. It really is. When supervisors evaluated employees, they accepted self-reported work 68% of the time when it was done without AI. But, and here is the kicker, when AI was involved in that exact same work, that trust plummeted to just 24%. Which is a massive drop.
[00:01:23] Yeah. And it is not just a vague feeling of skepticism from the boss, either. This translates directly into, you know, poor ratings of performance, lower perceived contributions, and just a massive hit to an employee's promotability. Right. Which brings us to our core question today. Conformal organizational frameworks, explicit policies, and transparent processes effectively neutralizes trust penalty and create a safe environment for AI integration. And I know exactly where you stand on this. Yes, you do. I am taking the position that they absolutely can.
[00:01:52] The penalty we are seeing right now is largely a symptom of policy silence. When organizations establish explicit legitimacy, well, we can override these human biases and establish what the literature calls calibrated trust. And I will be arguing the exact opposite. This trust penalty is rooted in fundamental, deeply ingrained human psychological biases. Structural policies simply cannot overcome them. Do you think we're just doomed?
[00:02:17] I think the penalty is driven by an unconscious effort heuristic and a really profound transparency dilemma. In my view, organizational mandates are entirely insufficient to prevent the inevitable cycle of hidden AI use and biased evaluations. Well, let me lay out exactly why institutional structure is the answer here. The AI trust penalty is thriving right now because it exists in a vacuum. I mean, we have companies buying massive enterprise licenses but leaving their managers entirely in the dark. In the dark about how to evaluate the workflow, you mean. Exactly.
[00:02:46] But there is a clear way out. The Myers findings from 2026 showed that when an organization explicitly encourages AI use, the trust penalty between AI users and non-users shrinks to non-significance. The gap just disappears. On paper, maybe? No, in practice. We can replace this current culture of confession-style disclosure, which feels to the employee like admitting a sin, frankly, with standardized process transparency. Take Judge Brantley Starr's standing order in the federal courts. The Texas judge, right? Right.
[00:03:16] He didn't ban AI. He simply mandated that any AI-drafted language must be verified by a human against traditional legal sources. By doing that, he shifted the entire focus away from how much manual effort the lawyer exerted. Which is great for a courtroom, sure. But the principle applies everywhere. He placed the focus squarely on the quality and accuracy of the verified outcome. When an institution writes the rules of engagement clearly like that, human bias can be structurally neutralized.
[00:03:42] I just, I come at this from a very different angle because the evidence suggests this penalty goes so much deeper than just a lack of clear guidelines. Deeper how? It is a profound, almost evolutionary psychological reaction. Consider the 2025 research by Reif and colleagues. They demonstrated what they call a social evaluation penalty. Right, where AI users are judged more harshly. Yes, where AI users are inherently perceived as lazier, less competent, and less diligent.
[00:04:10] A perfectly crafted policy memo from HR does not rewire human psychology. But... Hold on, let me finish. This brings us to the transparency dilemma identified by Shilke and Reimann. Through 13 different experiments, they prove that the mere act of disclosing AI use inherently erodes trust. Even when your company officially mandates it. Because it violates norms. Because it violates our prevailing norms of legitimacy. Exactly. It literally makes the work seem less proper in the brain of their reviewer.
[00:04:37] You can force employees to disclose their AI use, sure. But you cannot force managers to respect them for it. So what happens? It creates a vicious cycle. People engage in shadow AI, hiding their use entirely, and the honest employees who actually follow your structural policies are the ones who face undue scrutiny. Okay, but does it have to be that way? Look at what happens when a CEO completely shatters that scrutiny from the top down. You're talking about Shopify. Yes, Shopify.
[00:05:03] To understand if policies can solve this, we have to look at what actually happens when leadership decisively removes the stigma of cutting corners. In April 2025, Toby Letka issued an internal memo that didn't just passively permit AI use. It made reflexive AI use a baseline expectation. He basically told him to stop wasting time. Exactly. Teams literally had to prove why AI couldn't do the work before they were allowed to ask for additional headcount. Look at JPMorgan Chase rolling out their LLM suite.
[00:05:30] That's basically a secure, walled-off version of ChatGPT rolled out to roughly 200,000 employees. Right. Massive scale. When the organization explicitly normalizes the tool at that scale, it provides a safe harbor. It signals that AI use is highly compatible with seniority and good judgment. The Myers study proved that this kind of explicit encouragement neutralizes the penalty. The manager no longer views the AI use as a covert shortcut. But that assumes middle managers process top-down directives purely logically.
[00:05:59] And anyone who has ever worked in a corporation knows they absolutely don't. Oh, come on. They follow the metrics they're graded on. No, we have to look at the mechanics of the effort heuristic. For thousands of years of human history, value was directly correlated with physical caloric expenditure. Chopping wood, harvesting crops, it takes visible calories. Sure, manual labor. And when we transitioned to knowledge work, we couldn't see the calories being burned anymore.
[00:06:24] So our brains substitute at time spent and visible stress as proxies for value and dedication. An AI breaks that proxy. I agree with that. It completely shatters it. So even with top-down mandates from a CEO, evaluators unconsciously use that visible effort as a proxy for integrity. If an AI saves an employee three hours, the manager's brain inherently feels the employee lacks diligence. You really think we can't adapt? You cannot just mandate away a subconscious cognitive shortcut.
[00:06:53] Look at Slack's fall 2024 workforce index. They surveyed over 17,000 desk workers. Nearly half. 46% feared that using AI would make them look lazy or less competent. Well, yeah. People are anxious about new tech. If you've ever hesitated before telling your manager you used AI to outline a report, this is exactly why. You aren't being paranoid. That fear is an incredibly accurate read of how human managers actually evaluate work.
[00:07:17] You give 200,000 employees a tool, but if the middle manager feels cheated out of the sweat equity they expect to see, the penalty remains. You're completely right that it feels like a disruption of cognitive labor. But I think you're really underestimating how quickly institutional norms can override our egos. Let's look at historical precedent. Let me guess. Calculators. Yes, calculators. When electronic calculators were first introduced in finance and accounting, using one was literally seen as cheating.
[00:07:44] The visible effort of long division was the proxy for an accountant's diligence. If you didn't do the math by hand, well, you weren't a real accountant. Okay, but a calculator... Let me land the plane here. Eventually, the institution intervened. The sheer accuracy of the financial outcome entirely replaced the effort proxy. Why wouldn't AI follow that exact same normalization curve once explicitly legitimized? Because a calculator is deterministic. It still saves time and reduces manual effort. Yes, but it does basic, predictable arithmetic.
[00:08:13] Generative AI performs complex, creative, and analytic work. That is work that we attach deeply to human identity and profession and competence. It's just a different type of tool. A calculator doesn't write your strategic brief or design your marketing campaign, right? AI does. And that brings us to how this actually plays out in daily practice. You mentioned standardized process transparency earlier, but the data shows that forced disclosure backfires spectacularly. Does it, though? Yes. Atlassian's teamwork lab ran an experiment with almost 1,000 U.S. knowledge workers.
[00:08:43] They had them write an identical email. When the email explicitly disclosed that AI was used to help write it, the raters judged the author as substantially lazier. Okay, but by how much? Their chances of being recommended for a high-visibility project drop by 24 percentage points. 24 points for the exact same email. What is happening in the raters' brain there? They're adjusting for the perceived ease of the task. The moment they see the words generated by AI, their brain instantly discounts the cognitive labor.
[00:09:12] They think, well, this was cheap to produce, so it must be worth less. When an organization requires disclosure, it creates a trap. The honest employees disclose their use and get penalized. The savvy employees just hide it. Wait, let me make sure I'm wrapping my head around this transparency dilemma concept you're pushing. Are you saying that doing exactly what your boss tells you to do, disclosing the AI use as per a company policy, actually makes you look worse in their eyes? Exactly. Because it functions like a confession.
[00:09:39] Even if the rulebook says it is allowed, the psychological heuristic says you didn't put in the calories. You didn't do the work. Okay, but you are completely ignoring the alternative risk here, which the literature shows is far more catastrophic. Shoki and Ryman's own 2025 study proved that while voluntary self-disclosure might carry a small penalty, being exposed as an AI user by a third party damages trust far, far more severely. Well, sure, getting caught lying is always bad.
[00:10:05] Right. If you hide it and then your manager catches an AI hallucination or a weird formatting quirk, your credibility is entirely destroyed. So if third party exposure is the most damaging outcome possible, isn't standardizing process notes the only rational way to protect employees? In theory. In practice. Look at Kaiser Permanente's role out of ambient AI scribes in clinical settings. That's a medical setting. That's completely different. Just hear me out. For those who aren't familiar, this isn't a robot typing on a keyboard.
[00:10:33] It's a passive system listening to the doctor-patient conversation in the background, instantly structuring it into medical notes. The patient consents, the AI drafts the notes, the physician reviews and edits them, and the human remains entirely accountable. Okay, and? The verification is perfectly baked into the workflow. It's totally routine. When everyone discloses it as a standard operating procedure, it carries no stigma. It stops functioning as a confession of laziness and starts functioning as a mark of professional rigor.
[00:11:02] Kaiser Permanente is a highly specialized clinical environment with life-or-death compliance standards. In a standard corporate office, these standardized notes inevitably devolve into mere compliance rituals. Rituals that protect the worker. And who weaponizes those rituals? Managers. The manager sees the mandatory AI disclosure note on a project. Their effort heuristic gets triggered, and they unconsciously decide to scrutinize that honest employee's work twice as hard.
[00:11:28] The disclosure doesn't protect the employee. It paints a massive target on their back. I think you're being a little cynical about management. The real risk here isn't just that honest people get scrutinized. It's that this psychological pressure creates an unmanageable wave of shadow AI. KPMG ran a global study showing that 57% of employees hide their use of AI and present the work entirely as their own. 57%? 57%? 57%. When you mandate transparency in a culture that psychologically punishes it, you destroy organizational learning.
[00:11:58] Workflows that actually succeed can't be shared with the team, and much more importantly, errors can't be caught until it's too late. But the errors are exactly why we need structural process transparency. Look at the legal and reputational disasters we've already seen in the news. You mean the lawyers? Yes, Mata Villavianca. Lawyers were heavily fine for filing a brief with fictitious chat GPT citations because they didn't have a verification process. Or Deloitte Australia. They had to refund hundreds of thousands of dollars for an AI-generated report that completely fabricated a court judgment.
[00:12:29] Which proves my point about the danger of hidden AI. No, it proves the risk for a company is not in using AI. The risk is in unverified output. If we abandon structural transparency just because we're afraid of a psychological hurdle in middle management, we are inviting institutional disaster. I hear the institutional risk argument. I really do. But we need to talk about exactly who pays the price for this transparency on the ground. Because the burden of this trust penalty is highly uneven.
[00:12:57] And that is where structural policies completely fall apart. Uneven how? Policies are not enforced by robots. They are enforced by humans who carry intersecting cognitive biases. There was a fascinating 2025 study by Akar and colleagues looking specifically at software engineers reviewing identical Python code. Okay, I know this one. When reviewers believed AI was used, the competence ratings for the code were dropped by 9% across the board. But it wasn't distributed equally.
[00:13:24] Women and older engineers were penalized significantly more severely for the exact same AI-assisted code. Which is terrible, obviously. And to make it even more complicated, do you know who the harshest judges applying this penalty were? Senior male engineers who hadn't adopted AI themselves. Right. Those findings are critical to understand. But I firmly believe they point to a solution, not a surrender. Let me finish the thought on the behavioral outcome first. Because it is devastating.
[00:13:50] In that exact same company, only 31% of women engineers have even tried the AI coding assistant after 12 months, compared to 43% of men. Wow. The engineers who anticipated the greatest penalty from their managers essentially opted out of using the tool altogether. So your structural policy of explicit legitimacy, it completely fails marginalized groups. Because the managers enforcing it filter that policy through their own subjective biases about who gets the benefit of the doubt. There's a lot to unpack there, but I think we need to be careful.
[00:14:18] Are we conflating a general demographic bias with an AI-specific problem? They are inherently linked in this context. The fact that women and older workers are judged more harshly on subjective criteria is a well-documented, systemic issue that predates AI by decades. But that is exactly my point. By shifting evaluations to strictly outcome-based criteria, don't we systematically remove the subjective effort proxy that allows those demographic biases to manifest in the first place? That's a nice theory.
[00:14:48] If left to a manager's intuition, biases absolutely thrive. But if companies track the data, audit who bears the penalty, and intervene, they can fix it. Tracking the data just tells you the house is on fire. It doesn't actually put it out. But fluency training does. Unfamiliarity breeds suspicion. The Reif study showed something really important. Managers who use AI frequently themselves do not judge other AI users as lazier. Because they know how hard it is to actually get a good prompt. Exactly.
[00:15:17] The bias fades entirely with familiarity. Look at how Zapier handled this. They ran company-wide hackathons where employees in every single function had to build something with AI tools. Their CEO reported that weekly AI use jumped from under 10% to over 50% in a single week. A hackathon is a controlled environment, though. But what happens in the brain during that week? When a manager actually spends time building with AI, they realize you have to iterate. You have to prompt. You have to aggressively review the output.
[00:15:45] They realize the effort isn't gone. It's just shifted from drafting to editing. I don't deny that hands-on experience helps. Right. When you build manager fluency structurally, giving them hands-on practice, you dismantle the bias. The senior men in the Akar study were harsh judges precisely because they hadn't adopted the tool themselves. Mandated fluency and calibration training fixes that blind spot. Fluency training sounds great in a vacuum, but you are assuming managers have the time, the psychological safety, and the incentive to become fluent.
[00:16:15] They have to make the time. Let's look at Charter's October 2026 AI poll survey. Managers were 1.6 times more likely than rank-and-file workers to conceal their own AI use. Which again is a symptom of policy silence from the executives above them. No. It is a symptom of the transparency dilemma acting at every single level of the corporate hierarchy. Managers are caught in an impossible squeeze. They face top-down pressure to show massive AI-driven productivity gains,
[00:16:42] while battling their own deep-seated uncertainty about what counts as legitimate effort. So they just hide it. Exactly. They use AI quietly to hit their metrics, and then they judge their own teams harshly for doing the same, because they unconsciously project their own feelings of cheating onto their subordinates. You cannot train away a psychological conflict of that magnitude with a one-week hackathon. I disagree. If the managers themselves feel they have to hide their use to maintain their authority, how can they possibly evaluate their team's disclosure objectively?
[00:17:09] They evaluate it objectively when the organization demands outcome-based criteria. This is where we have to translate the state into actionable steps for anyone managing a team today. You have to define what good work looks like purely in terms of accuracy, impact, and timeliness. Easier said than done. You evaluate the judgment the employee used in verifying the AI's output. You stop measuring crude effort metrics entirely. We are seeing Microsoft and Mena both make moves to explicitly credit AI-driven impact in their performance reviews.
[00:17:39] But let me finish this point. When the institutional machinery, the actual rubrics, the promotion criteria, the talent selection panels, is redesigned to reward the verified outcome, the manager's personal psychological discomfort becomes irrelevant. The system overrides the bias. But the system is fundamentally made of people. You can change the rubric on the HR form all you want, but a manager who fundamentally believes an employee is cutting corners will find a way to reflect that on the performance review. They can't if the metrics are objective. Oh, please.
[00:18:09] If they can't penalize them for effort, they'll say the employee lacks leadership presence or independent initiative. The laziness heuristic doesn't disappear. It just mutates into a different subjective penalty. Which is exactly why my final point is about achieving calibrated trust. In the human automation literature, calibrated trust means relying on a system exactly as much as its actual capabilities warrant. Right. Not over-trusting or under-trusting. Exactly. Not blind trust where we accept hallucinations,
[00:18:37] and not reflexive distrust where we punish efficiency. By establishing clear policies, naming approved tools, requiring standard process notes for verification, and holding managers accountable through audits, we build a culture where trust is tied to the reliability of the work. I just don't think it's that simple. If you're a mid-level manager trying to implement this tomorrow, it means creating a tangible receipt of cognitive effort. A checklist of what the AI drafted versus what the human verified.
[00:19:05] Yes, human effort heuristics are deeply ingrained, but they only dictate behavior when there are no rules. Institutional design is the only way we separate the massive value of human judgment from our primitive anxieties over AI assistance. And I remain convinced that the AI trust penalty is fundamentally psychological. Not structural. Policies are necessary, certainly, but they are a surface-level fix for a deep-seated human condition. You really think we can't overcome it? You cannot easily legislate away the transparency dilemma.
[00:19:35] Nor can a CEO memo magically erase the disproportionate social evaluation penalties placed on different demographics in the workplace. The Reif data, the Atlassian experiment showing that 24-point drop, the Slack index, They all point to the exact same truth. Which is? As long as human evaluators unconsciously equate struggle with competence, the instinct to conceal AI use will remain a persistent, highly rational barrier in the workplace. Mandates don't change human nature. They just change how well we hide it.
[00:20:04] Well, I think we have reached a point of mutual convergence on one crucial fact, at least. Policy silence is the most dangerous path an organization can take right now. Oh, absolutely. When companies say nothing, they leave their employees completely unprotected, adrift in a sea of unchecked biases. Completely agreed. Silence just lets the worst of our psychological heuristics run rampant. But we certainly maintain a very substantive disagreement on whether institutional redesign
[00:20:31] is actually powerful enough to overcome the effort heuristic entirely. Fair enough. And that shifting dynamic of human automation interaction is something that is playing out in offices around the world right now. It offers a rich area for observation in your own workplaces, leaving you to form your own conclusion. Will we learn to value the precision of the factory-built chair? Or will we always demand to see the calluses on the carpenter's hands? It really is the defining workplace question of our time. We'll see you next time.


