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] So picture the scenario. It's Friday afternoon. The best time of the week. Right, exactly. You're a manager and, boy, frankly, you are just exhausted. The week has been a total marathon. We've all been there. Yeah. And you have two analysts on your team. They both hand in this really crucial market summary right at the four o'clock deadline. Okay. Down to the wire? Right. So you pull up the files, you read them over, and I mean, they are both absolutely flawless. Oh, wow. That's rare.

[00:00:30] Totally. The data is accurate, the strategic insights are super sharp, the formatting is impeccable, and they hit the deadline. Now, analyst A drops by your desk, hands you the report, and just sort of casually mentions that she used the company's approved AI assistant to draft the initial framework. Okay. Totally normal these days. Right. And she explains how she rigorously went through and manually verified every single figure, every citation, all the strategic claims. So she's showing her work. Exactly.

[00:00:56] Exactly. Now, analyst B hands in the exact same quality of work, hits the exact same level of insight, but says absolutely nothing about how it was produced. Interesting. Yeah. So you smile, take the reports, head home for the weekend. But Monday morning rolls around, you walk into the office, and your director tells you, hey, you need to pick one of these two analysts for a highly sensitive high-stakes briefing with the board of directors.

[00:01:19] Oh, man. High-stakes indeed. So I want you, the listener, to just sit with this for a second, put yourself in that manager's chair. Who do you instinctively want to give that assignment to? See, it creates this really fascinating internal tension, right? Because the logical, rational part of your brain immediately votes for transparency. Yeah. You want the person who documented their workflow.

[00:01:42] You'd think so. You want analyst A, right? Because she was honest, she was incredibly efficient, and she explicitly followed the verification protocol. Right. But here is the incredibly uncomfortable truth that we are exploring today. Analyst A, the honest, communicative one, she's just put herself at a staggering disadvantage. Wait, really? By just being honest?

[00:02:06] Yes. By simply telling the truth about her highly efficient process, she has unknowingly triggered this massive psychological trapdoor in your brain. That is wild. And, I mean, this trapdoor is creating a profound level of friction in the modern workplace right now. Oh, absolutely. We're currently watching this huge collision between sweeping corporate mandates and fundamental human psychology. Like, on one side, you have executive leadership practically demanding everyone adopt these expensive new tools to drive productivity.

[00:02:35] Right. The top-down pressure is immense. Yeah. And on the other side, you have this deeply ingrained, almost primitive human instinct whispering to middle managers like, wait a minute, is using that software just a sophisticated form of cheating? Exactly. And to figure out how this collision is actually playing out on the ground, we're diving deep into a brilliant and honestly somewhat techifying new paper. Yeah. It's called The AI Trust Penalty, authored by Jonathan H. Westover. It's a fantastic piece of research.

[00:03:01] It really is. And we are recording this deep dive on Wednesday, November 18th, 2026. Let me just set the stakes for you right now, whether you're commuting at the gym or sitting at your desk. It affects literally everyone in knowledge work. Totally. Whether you're managing a global team, angling for a promotion this quarter, or honestly just trying to get through your daily workflow without accidentally sabotaging your own reputation. Understanding this hidden tax on honesty is absolutely critical to your career survival.

[00:03:27] Survival metric is really the perfect phrase for it because, you know, this penalty is entirely invisible. Right. You will never see it on a formal performance rubric. Human resources doesn't have a little checkbox for it. Yeah. There's no box for used AI too much. Exactly. It operates entirely in the shadows, governing how human beings subconsciously evaluate each other's character, their dedication, and ultimately their worth to an organization. Okay. Let's unpack this because we need to start by looking at the actual numbers.

[00:03:56] And they paint a very stark picture of what happens when people are honest. The data is pretty undeniable. Yeah. The paper heavily features this behavioral study published recently. And the researchers wanted to isolate exactly what happens to a manager's trust when an employee simply admits to using AI. Right. Just the admission itself. Yeah. So they designed this experiment where supervisors had to make a choice.

[00:04:22] Either accept an employee's self-reported work at face value or actively choose to verify it. And the methodology here is what makes the findings so robust. I mean, we really have to differentiate between survey trust and behavioral trust. Right. Because surveys can be a little fluffy. Exactly. If you just send a survey to a manager asking, Hey, how much do you trust your team? You're going to get highly aspirational answers. Right. People want to sound like good managers.

[00:04:49] Yeah. People answer surveys based on who they want to be. Like what they think the socially acceptable answer is. They say, Oh, of course I trust my team. I'm this empowering leader. But this study didn't do that. No, it didn't. It measured behavioral trust by forcing the supervisors to pay a tangible cost. Meaning they had to put their money where their mouth is, essentially. Precisely. They were given a finite budget of resources, which, you know, represented time or social capital. Okay. And they had the option to spend those resources to double check the employee's work. Wow.

[00:05:20] So it wasn't just a free check. Right. This elevates the scenario from just a vague feeling of skepticism to actionable, measurable distrust. That makes so much sense. The manager in this simulation is effectively saying, My confidence in you is so remarkably low right now that I am willing to penalize my own limited budget just to make sure you aren't pulling one over on me. Which makes the contrast in their data absolutely wild. I mean, let's look at the baseline first. The baseline is fascinating. Yeah.

[00:05:48] When the supervisors were under the impression that the work was completed without any AI assistance, they accepted the employee's self-reported work without spending any capital to verify it 68% of the time. So well over two-thirds of the time. Right. The default human setting was basically, It looks good. I trust you. Let's move on. But the moment they were informed that AI was involved in the drafting process, that acceptance rate plummeted to 24%. It's just a catastrophic collapse of faith.

[00:06:16] I mean, you go from the vast majority of work being waved through on this assumption of competence to nearly 80% of it being treated as a potential liability. And if you're listening to this and thinking, Okay, so my boss double checks my spreadsheet. What's the big deal? Oh, it's a huge deal. Right. Because the fallout didn't stop at that single isolated task. It wasn't a scenario where the manager audited the document, found it was perfectly fine, and then just went back to treating the employee normally. Not at all. The paper points out that this specific instance of distrust acted like a virus.

[00:06:46] It literally infected the manager's entire holistic view of that person. And that contagion effect is arguably the most alarming takeaway for anyone trying to build a career right now. Yeah, it's terrifying. The supervisors who chose to penalize the AI use didn't compartmentalize it. They didn't think, oh, I just don't trust this specific weekly report. Right. That seed of doubt bled over into how they evaluated the human being as a professional entity entirely. Wow.

[00:07:11] The data showed this disclosure led directly to significantly poorer ratings and overall performance reviews. Because they just view you differently now. Exactly. It fundamentally lowered the manager's perception of the employee's long-term contribution to the organization. And, I mean, cutting right to the chase for anyone listening, it severely downgraded the employee's perceived promotability. You know, it reminds me of a very specific feeling from childhood. Oh, yeah?

[00:07:38] Yeah, this feels exactly like when scientific calculators were first introduced in middle school math classes. Oh, that is a perfect example. Right? You could get the right answer on the test. It could be perfectly accurate. But if you didn't show the manual labor, like all the messy, agonizing scribbles in the margins. The teacher assumed you didn't actually understand the underlying principles of the math. Exactly. You were suspected of taking a shortcut that bypassed actual learning.

[00:08:04] And it feels like we are applying that exact same standard to professional adults right now. We absolutely are. We're grading knowledge workers on showing their work, even when the final deliverable is indistinguishable from, or like even better than, a purely manual effort. That analogy perfectly captures the underlying theoretical framework that organizational psychologists use to understand how trust is built and how it's broken. Right.

[00:08:31] Back in 1995, Mayer and his colleagues published this really seminal piece of literature. They called it the integrative model of trust. Okay, the integrative model. Right. It basically posits that trust isn't just a single emotion. Trust is our willingness to make ourselves vulnerable to another person. That makes sense. And that willingness relies on a three-legged stool. Ability, benevolence, and integrity. Ability, benevolence, integrity. Okay.

[00:09:00] So if I walk into a one-on-one with my boss and I just cheerfully admit that I used a language model to synthesize a massive client portfolio, how am I inadvertently kicking the legs out from under that stool? Let's look at it through the subconscious lens of the manager who is receiving that news. Okay. So the first pillar is ability.

[00:09:19] The moment you declare that an algorithm did the heavy lifting on the first draft, the manager's brain quietly asks, wait, could you have actually synthesized this portfolio without the machine? Oh, wow. So they doubt your actual skills. Right. Do you genuinely possess the analytical skills I hired you for, or are you just like a good prompt engineer? Because they literally can't untangle where the machine's competence ends and mind begins. Exactly. So the perception of your inherent ability takes a hit.

[00:09:49] Then we move to the second pillar, which is integrity. Okay. The manager thinks, well, you finished this comprehensive synthesis in two hours, but you are salaried for a 40-hour week. Oh, I see where this is going. Yeah. Are you obscuring how much effort this job actually requires? Are you lying about your bandwidth? That is brutal. It severely damages the perception of your integrity. And finally, you have the third pillar, benevolence. Which is what exactly?

[00:10:18] Benevolence is the belief that the other person has your best interests at heart. Okay. So the manager's instinctual reaction might be, are you prioritizing your own convenience over the rigorous quality control our clients expect? Wow. Like, are you gaming this system to make your own life easier while leaving me holding the bag if this thing hallucinates? That is just a devastating triple threat to a professional reputation. It really is. And the paper makes a really crucial distinction here that I think we need to clarify. Yes.

[00:10:46] The difference with algorithm aversion. Right. The author firmly separates this specific AI trust penalty from a broader, older concept called algorithm aversion. It is so vital to separate the two. Because, you know, algorithm aversion has been a subject of study for over a decade now. It's basically when people just don't like AI, right? Exactly. It describes the human tendency to deeply distrust a piece of technology itself. Particularly after witnessing it make like a single mistake. Right.

[00:11:15] We hold machines to this impossibly perfect standard. I mean, think about it. If a human intern makes a typo, we forgive them. Sure. Humans make mistakes. But if a GPS system routes us into a traffic jam one time, we declare the entire system useless and refuse to trust it for the rest of the road trip. That is so true. That is algorithm aversion. But the AI trust penalty operates on an entirely different psychological axis. Because the trust penalty isn't about the software at all. It's personal. It is deeply personal.

[00:11:46] Algorithm aversion means I don't trust the tool. The AI trust penalty means I no longer trust you because you chose to pick up the tool. Wow. And the research base, particularly this robust set of studies conducted in 2025, they actually controlled for people's natural baseline skepticism of algorithms. Oh, really? So they factored that out? Yes. They found that even when a manager was perfectly comfortable with the technology itself, the penalty against the employee remained. That is wild.

[00:12:15] The manager isn't evaluating the software's code at that point. We are evaluating the employee's character. Which brings up such an interesting paradox. Yeah. I mean, if managers are silently penalizing their own direct reports for finding faster, more efficient ways to execute their jobs, it forces us to ask a really fundamental question about the modern workplace. Right. Why are we like this? Exactly. Why is human psychology wired this way?

[00:12:41] Why do we look at a technological leap that saves hours of drudgery and instantly categorize the person using it as having a character flaw? To get to the root of that why, we really have to examine how human beings have been conditioned to measure dedication. And this leads us to a fascinating large-scale behavioral study from 2025 that ran four pre-registered experiments with nearly 4,500 participants. That's a huge sample size. It is.

[00:13:06] The goal was to isolate the exact personality trait that evaluators were subconsciously mapping onto employees who used AI. So they were looking for the specific label people were quietly applying to these workers. And the label they found was incredibly consistent. It was laziness. Laziness? Yes. Wow. Yes.

[00:13:27] The experiments revealed that employees who received assistance from generative AI were universally rated as lazier, less competent, and less diligent than employees who received the exact same amount of assistance, but from traditional non-AI sources. Wait, hold on. So if I ask a human colleague on the marketing team to help me brainstorm five headlines, I'm seen as a collaborative team player.

[00:13:49] But if I ask an LLM to generate five headlines, I'm perceived as lazy, even though the time saved and the output might be completely identical. That is exactly what the data demonstrates. The source of the assistance entirely changes the moral judgment applied to the worker. That's so irrational. It is. And this dynamic is further unpacked by another extensive body of research detailing what academics call the transparency dilemma. The transparency dilemma.

[00:14:16] Through a series of 13 different experiments, researchers pinpointed the psychological mechanism driving this judgment. They call it the effort heuristic. Okay, so a heuristic is basically a mental shortcut, right? Because managers don't have the time or the capacity to audit every single keystroke an employee makes, they need a proxy. They need a shorthand way to measure if someone is, you know, a good worker. You hit the nail on the head. The human brain processes an overwhelming amount of social information every single day. Oh, yeah.

[00:14:45] And to survive that cognitive load, we rely on these heuristics to make snap judgments. The effort heuristic is our deep-seated, largely unconscious tendency to use visible struggle, time spent, and literal sweat as a proxy for value, loyalty, and dedication. So if I walk past your desk at 7 p.m. and you are staring intensely at your monitor, rubbing your temples and surrounded by empty coffee cups, my brain immediately logs a data point. Right.

[00:15:14] My brain says, that person cares deeply about this company. That person is highly dedicated. Exactly. We have spent centuries, honestly, dating back to the Industrial Revolution, building workplace cultures around the visible performance of labor. Like punching a clock. Exactly. You punched a clock. Your value is directly tied to your physical presence and the visible exertion of energy on an assembly line. And we've just brought that into the office.

[00:15:40] We have carried that exact same industrial-era architecture right into modern knowledge work. Wow. But then, generative AI enters the ecosystem and completely obliterates the visible struggle. The data analysis that used to take five agonizing hours of spreadsheet formatting now takes 30 seamless seconds of prompting. And so the manager's mental proxy for dedication, the visible effort, the temple rubbing, is suddenly absent. Right. Your heuristic breaks down completely.

[00:16:09] The manager's brain does not know how to compute this lack of struggle as high efficiency. So it defaults to computing it as a lack of integrity. Exactly. It computes it as laziness. But I have to push back on this a little bit because it feels completely irrational when you look at the fundamental purpose of a business. How so? Well, we operate in a capitalist structure that is literally built on the concept of maximizing output while minimizing input.

[00:16:35] The entire goal of corporate innovation is to reduce the cost of production. Very true. So why on earth do we still equate suffering and sweat with value in the year 2026? Shouldn't a rational manager be absolutely thrilled? I mean, shouldn't they be throwing a parade that an employee accomplished a five-hour task in 30 minutes? Logically, yes. Right, because that is four and a half hours of freed up human capital that can now be deployed toward higher level strategic thinking. Your logic is entirely sound from an economic perspective.

[00:17:04] But it assumes that organizational culture evolves at the exact same speed as technological innovation. And it doesn't. No. History tells us that is never the case. Technology can transform overnight with a software update. Human culture takes generations of unlearning to shift. That's a really good point. We are dealing with a severe cultural lag. I mean, think about the transition from typewriters to word processors. Oh, that's a good comparison.

[00:17:30] When word processors first entered the office, early adopters were sometimes viewed with suspicion by traditionalists. Wait, really? Yeah. If your typed page didn't have whiteout on it, if you didn't have to manually retype an entire page because of one error, there was this lingering feeling that you hadn't really worked as hard as the person sweating over the ink ribbon. That is so funny to think about now. Right. But we are experiencing that exact same friction today, just on a much more profound scale. That makes a lot of sense.

[00:17:58] The technology outpaces the etiquette. Exactly. And that lack of etiquette brings us to another critical factor identified in the research, which is the legitimacy mechanism. How does legitimacy play into a manager thinking I'm lazy? It has to do with the fact that the norms surrounding AI in the workplace are currently in a state of total chaos. Yeah, nobody knows what the rules are. Right. There is no universally agreed upon standard operating procedure. So when an employee voluntarily steps forward and discloses their AI used to a supervisor,

[00:18:28] because the cultural norms are so muddy, the mere act of disclosure makes the work feel improper. Oh, I see. It feels akin to a confession. You are confessing to a behavior that hasn't officially been written into the corporate rulebook as a fireable offense, but it still carries an illicit taboo energy. You're putting the manager in this weird position where they have to decide on the spot if what you did was okay. Exactly. And because it feels secretive, their default is suspicion. Right.

[00:18:55] And it gets even more precarious if you choose not to confess, but you get caught anyway, right? The research highlighted a terrifying finding for anyone trying to fly under the radar. Oh, the third-party exposure data? Yeah. Being exposed as an AI user by a third party destroys trust even more severely than self-disclosure. It creates the ultimate professional double bind. Option A. You tell the truth proactively,

[00:19:20] and you suffer a penalty because your manager's effort heuristic flags you as lazy and your process as illegitimate. Lose-lose. Right. Option B, you stay completely quiet, but someone else exposes you. Maybe the IT department runs an audit on enterprise software usage or a colleague casually mentions your workflow in a meeting. Which happens all the time. The trust destruction in Option B is catastrophic. Because now you aren't just viewed as a lazy employee looking for a shortcut.

[00:19:49] You're viewed as an actively deceptive employee hiding things from leadership. So, put yourself in the shoes of the listener right now. Now, if you are an employee sitting at your desk today and you know for a fact that using this technology makes you incredibly fast, highly accurate, and less stressed. Which it does. Right. But you also know that telling your boss about it makes you look lazy and getting caught makes you look like a liar. What is the logical move?

[00:20:16] You do the only rational thing a human being can do when trapped in a contradictory incentive structure. You lie. Yep. You use the tool to reap the benefits of efficiency and you hide the evidence to protect your reputation. And that incredibly rational decision leads us to the next staggering realization in the paper. The concealment epidemic. It's massive. We are essentially watching the birth of a colossal global shadow AI economy.

[00:20:43] And it is operating right inside our own offices completely undetected by the people supposedly running the companies. And to understand the gravity of this shadow economy, we first really have to look at the sheer scale of adoption. Okay. Let's look at the numbers. The speed at which these tools have saturated the knowledge work sector is truly unprecedented in the history of enterprise technology. The numbers are literally mind boggling. They are.

[00:21:07] The joint global survey published back in early 2024 showed that 75% of knowledge workers were already using AI in their daily workflows. That was 2024. Right. And by the spring of 2026, subsequent tracking data found that a staggering 94% of U.S. knowledge workers were utilizing it. 94%. That's practically everyone. We are talking about practically every single person with a laptop and a Wi-Fi connection. It is ubiquitous. But the true story isn't the adoption rate. What is it then?

[00:21:37] The true story is the chasm between adoption and transparency. When you contrast those massive usage numbers with the concealment statistics, a very dark picture emerges. Okay. Lay it on me. In that same foundational 2024 survey, 52% of users explicitly stated they were reluctant to admit they used AI for their most important high-stakes tasks. 52%? Right.

[00:22:03] And a subsequent survey focused on workplace communication found that 48% of employees were deeply uncomfortable discussing their AI use with their direct managers. So they are fine using it for trivial things, but the moment it touches core competency, they just go silent. Exactly. And the most illuminating data point comes from a 2025 global study spanning 47 countries. Oh, wow. So it's a global phenomenon. Completely.

[00:22:27] They found that 57% of employees are not just reluctant to talk about it, they are actively, intentionally hiding their AI use and passing the digitally generated work off entirely as their own manual creation. Over half of the global professional workforce is actively running a covert intelligence operation against their own management structures. That's one way to put it, yeah. That is an astonishing reality to wake up to. And as we discussed, it is a completely rational, defensive response. It is.

[00:22:58] I mean, think about it. If your CEO goes on a financial news network and passionately declares, you know, our company rewards AI innovation, we are building the future. But your direct middle management supervisor quietly docks your quarterly performance review because you didn't visually suffer enough to produce the latest sprint report. Exactly. You are going to listen to the supervisor. You are going to use the tool in absolute secret. It is basic behavioral economics at play. Human beings respond to the immediate incentives in their local environment. Right.

[00:23:28] Whatever affects their day to day. If transparency is actively punished in the microculture of a team, transparency will rapidly disappear, regardless of what the macroculture of the corporation claims to value. That makes total sense. But there is a twist in the source material regarding this concealment that completely upends the traditional narrative. What's the twist? It is not just junior analysts or entry-level employees hiding their workflows from the big bad bosses. The managers themselves are hiding it at an even higher rate. Yes.

[00:23:56] The 2026 Pulse research data on this blew my mind. It's incredible, isn't it? The study found that managers are actually 1.6 times more likely to conceal their own AI use than the subordinates reporting to them. I mean, why on earth are the people with the authority hiding it more? They are the ones setting the team's culture. Because they are caught in an even tighter, more unforgiving psychological vice. How so? Think about the immense, multi-directional pressure placed on a middle manager today.

[00:24:22] From above, the executive suite is pounding the table demanding aggressive AI-driven productivity gains. Right. The do more with less mandate. The mandate is transform our operational efficiency. But from below, their team of direct reports is looking up to them for mentorship, guidance, and domain expertise. They have to be the wise elder. Exactly. But internally, sitting alone at their desk, the manager is likely grappling with a severe case of imposter syndrome.

[00:24:50] They're having an existential crisis about their own utility. Precisely. They are looking at the output of these models and quietly asking themselves, well, if an algorithm can write my quarterly strategic review, draft my delicate performance feedback emails, and balance my departmental budget in seconds, what exactly is my unique value proposition? Like, why am I here? Right. What is the company actually paying me a six-figure salary to do? So the manager uses the tool in secret to meet the crushing demands of the executive team. Right.

[00:25:19] But they hide that usage from their direct reports to protect their fragile ego and maintain their perceived authority and expertise over the team. It is just a cycle of mutual deception. And this widespread systemic secrecy is compounding an internal psychological burden that behavioral researchers have identified as the AI ghostwriter effect. The ghostwriter effect. That phrasing implies that someone else is doing the heavy lifting and you are just slapping your name on the cover. That is the exact sentiment.

[00:25:48] Researchers examining this phenomenon found that when professionals write complex documents or code with heavy assistance from generative models, they fail to develop a sense of psychological ownership over the resulting product. Oh, that makes sense. They feel alienated from their own output. They read a brilliantly structured memo that they prompted into existence and they think, well, I didn't really write this. Right.

[00:26:12] But because of the trust penalty, despite feeling like they didn't really author the work, they still self-declare as the sole creators. They absorb the praise, but internally they feel disconnected from the achievement. Have you ever experienced this? I mean, if you are listening right now, have you ever looked at a perfectly formatted and highly persuasive email you just prompted, hit send, received a compliment on it, and just felt a weird, hollow sense of detachment? It's a very common feeling right now.

[00:26:40] You have a workforce that is technically more efficient than ever before, but internally they feel like total frauds. Exactly. They are deeply ambivalent about their own output. They are carrying the chronic stress of maintaining a secret. And they are professionally isolated because they can't have an honest conversation with their peers about how they are actually accomplishing their daily tasks. Exactly. And that deep, pervasive isolation is where the true existential danger to the business actually begins. What do you mean?

[00:27:08] If absolutely everyone in the building is using this powerful technology, but absolutely no one is willing to talk about it on the record, what happens when the machine inevitably makes a mistake? Oh, which brings us to the fallout. Yes. We have to look at the cascading business risks, the highly public legal disasters, and the wildly unfair burdens this invisible penalty places on specific demographics within the workforce. It gets pretty ugly. Let's start with what the paper calls the destruction of organizational learning.

[00:27:38] Because to me, this seems like a fatal flaw for any company attempting to remain competitive in the next decade. It is an absolute silent killer for corporate growth and innovation. Think about how a healthy organization improves. Someone figures out a better way to do something, right? Exactly. Someone figures out a better way, they share it, and the whole company levels up. Right. But under the shadow AI economy, if a mid-level employee discovers a brilliant, highly complex prompting sequence

[00:28:05] that cuts client data processing time in half, they keep it a closely guarded secret. Because they're terrified of the trust penalty. Exactly. Therefore, the organization cannot scale that success. The massive efficiency wind remains permanently isolated in one single cubicle. The company, as a whole entity, learns nothing. You literally cannot build training modules around a workflow that officially does not exist. Exactly. And the flip side of that coin is even more dangerous.

[00:28:32] When covert workflows fail, they cannot be corrected. Because no one knows they're happening. We all know that these models hallucinate. We know they confidently invent facts and make logic errors. Sure. Industry surveys consistently note that over half of employees admit they have submitted work containing mistakes generated by AI. Half? Wow. Now, if the use of the tool is kept hidden from management, the supervisor cannot provide necessary oversight. They cannot implement guardrails or safety checklists.

[00:29:01] And they cannot help the employee learn how to prompt more effectively to avoid the error next time. So the mistake just gets buried. Or glossed over. Or worse, passed directly along to the end client. And when those hallucinated mistakes get passed along to a client or, say, a federal judge, we see these massive, intensely public disasters that make global headlines. Oh, the legal cases. The source material brings up the now infamous Mata v. Avianca case from 2023.

[00:29:30] This was the federal lawsuit in New York where two seasoned lawyers and their law firm were fined $5,000 and publicly humiliated. It was a huge story. Why? Because they filed a legal brief that cited completely fictitious, hallucinated court opinions that were hallucinated by ChatGPT. Fake case law. And it gets worse. When the opposing counsel and the judge couldn't find the cases and called them on it, the lawyers doubled down and actually tried to defend the fake cases. It was a watershed moment that sent a shockwave through the legal tech industry.

[00:29:59] It demonstrated the sheer confidence with which these models can fabricate reality. Yeah, it's terrifying. And then you look at the corporate sector with the Deloitte-Australia incident from late 2025. Oh, what happened there? Deloitte, a premier global consultancy, was forced to agree to partially refund the Australian government for a major contract worth hundreds of thousands of dollars. Yikes. The reason?

[00:30:22] The official report they delivered contained fabricated quotes attributed to a real court judgment alongside references to academic papers that literally did not exist outside the AI's imagination. They just made it up. They had utilized generative tools to speed up the research, and the tools had simply invented the evidence. But you pointed something out to me when we were reviewing this case law that completely shifted my perspective on these disasters. What was that?

[00:30:47] The ultimate, most crucial lesson to extract from these highly publicized failures isn't that using AI is inherently bad or dangerous. No, not at all. That is the wrong takeaway entirely. In the Avianca case, the presiding judge explicitly noted in his ruling that utilizing artificial intelligence to assist in legal research is not inherently improper. The ethical violation was not the use of the tool. The ethical violation was the failure to verify the output.

[00:31:16] That is such a huge distinction. The existing rules of professional conduct already placed a strict gatekeeping duty on the lawyers to ensure the accuracy of their submissions. And they completely abdicated that human duty to a machine. And this brings us right back to the vicious cycle of the trust penalty. Exactly. If employees are terrified of looking lazy, they conceal their AI use. If they conceal their AI use, they are significantly less likely to build rigorous, visible verification processes. Because it leaves a paper trail.

[00:31:45] Because if you have a highly visible, documented process for checking AI output, you are implicitly admitting that you used AI in the first place. The act of concealment practically guarantees that the output will remain unverified until it blows up spectacularly in the company's face. It fosters a toxic culture of plausible deniability, which is the exact opposite of what you need right now, which is a culture of radical accountability. Right. And we really cannot ignore the human cost of this dynamic either.

[00:32:15] Maintaining this facade requires immense effort. It sounds exhausting. Research tracking employee well-being found a direct, measurable correlation between hiding AI use and significantly higher odds of severe professional burnout. I believe it. But going into work every single day and actively hiding the reality of how you accomplish your tasks is incredibly cognitively taxing. It just drains your mental reserves.

[00:32:42] But the fallout of this penalty isn't just about heightened stress levels or the risk of legal fines. We have to talk about who actually bears the brunt of this punishment. This part is truly upsetting. The paper details an incredible, highly revealing study conducted in 2025. Focusing on the tech sector, researchers studied over a thousand software engineers. These engineers were asked to review identical snippets of Python code.

[00:33:07] The twist was, if the reviewer was led to believe that AI was used to write the code, they automatically slapped a 9% competence penalty on the coder's evaluation. 9%? Just off the top. But that penalty wasn't distributed equally across the workforce, was it? No, it was deeply, systematically unequal. The data revealed a stark demographic divide. Women and older workers were penalized the absolute most severely for utilizing AI assistance.

[00:33:37] That is so messed up. And when the researchers looked at who was driving this disproportionate punishment, they found it was largely dispensed by senior male engineers who notably did not use AI tools themselves. That is just infuriating. Think about the power dynamics at play there. It's very pronounced.

[00:33:55] You have the people with the most entrenched institutional power who are perhaps stubbornly refusing to adapt to the new technological reality, actively using this invisible trust penalty to punish marginalized groups who are proactively trying to innovate and stay efficient. And the downstream consequences of that dynamic are severe. It creates a chilling effect on diversity and technology adoption. Because why would they adopt it if they're just going to get punished? Precisely.

[00:34:20] Because the female engineers in this particular firm astutely anticipated that they would be judged more harshly than their male counterparts for using the exact same tools, they actively avoided adopting the technology. Wow. The longitudinal data showed that after 12 months, only 31% of the female engineers had integrated the AI coding assistant into their workflow, compared to 43% of the male engineers. That is a massive gap. So this penalty isn't just a matter of bruised reputations.

[00:34:47] It is creating a very real, measurable adoption gap. If left unchecked, this dynamic will drastically compound existing inequalities in career advancement, skill development, and earning potential over the next decade. Here is where the macroeconomics of this entire situation get really interesting to me. Okay. When you zoom out and look at the corporate balance sheets, this psychological trust penalty is effectively taxing the exact behavior that companies are spending hundreds of millions of dollars to encourage.

[00:35:16] The irony is rich. The paper mentions JPMorgan Chase as an example. They invested heavily to roll out their proprietary LLM suite to almost their entire workforce. We are talking about roughly 200,000 employees actively using it by mid-2025. That is a staggering rollout. I want you to just imagine the astronomical enterprise licensing and server costs associated with that rollout.

[00:35:40] And yet if a massive institution like that doesn't actively fix the microculture on the ground, you have a bizarre situation where the bank is spending a fortune on cutting edge tools while middle management is secretly punishing the employees who are actually brave enough to use them. It is a tragic, incredibly expensive irony. It is organizational self-sabotage functioning on a massive systemic scale.

[00:36:04] You are effectively buying a fleet of high-performance vehicles to help your team win the race, but your mid-level managers are hiding in the parking lot aggressively handing out speeding tickets to anyone who actually puts the car in drive. What a great analogy. The investment is completely nullified by the culture. Okay, we have painted a pretty grim dystopian picture of the current workplace landscape, but the incredibly valuable part of the paper is that it doesn't just admire the problem and leave us feeling helpless. No, it offers real solutions.

[00:36:34] Right. He synthesizes the research to provide a clear, actionable blueprint for how to actually fix this. He lays out five evidence-based organizational responses, essentially a playbook for breaking the cycle of shadow AI. And these are crucial. I want to spend some real dedicated time here because if you are a listener who manages a team, or even if you are just trying to manage your own career trajectory, this is the tactical roll-up-your-sleeves advice you need for tomorrow morning. Absolutely. Where does a company even begin to dismantle this penalty?

[00:37:02] If the root cause of the penalty is that AI use feels like an illicit shortcut because the rules are undefined, the absolute first lever a company must pull is to explicitly change the rules of the game. You have to make the technology unequivocally legitimate. You have to legalize it, essentially. Precisely. Take it out of the gray market. Exactly. Okay. Correct. And this isn't just theory. It is the strongest single intervention found in the entire body of research. Really? Yeah. Just saying it's okay works that well.

[00:37:31] Let's go back to that behavioral study where managed trust plummeted to 24% when they found out AI was used. The one with the finite budget to check work? Exactly. Yep. The researchers ran another variation of that exact same experiment. In the second variation, they established a condition where the organization explicitly, loudly encouraged the use of AI tools. Okay. Under those conditions, the trust penalty practically vanished. It shrank to the point of statistical non-significance. You're kidding.

[00:38:01] Supervisors in that environment trusted AI users 47% of the time, which was statistically indistinguishable from the baseline trust rate for non-users. Wait, let me make sure I'm getting this. Just by the company putting out a clear mandate saying, we want you doing this, the managers completely stopped viewing the behavior as lazy. Yes. Because clear executive communication neutralizes the legitimacy mechanism. Wow.

[00:38:25] If the CEO firmly states, using these specific AI tools is the new standard operating procedure for how we do business, then the employee is no longer breaking an unwritten taboo rule. They are simply following the new official rule. That makes total sense. Right. When there is silence from the top, it creates a leadership vacuum. And managers inevitably fill that vacuum with their old effort-based biases. So they just fall back on old habits. Exactly. Yeah.

[00:38:52] Companies need affirmative, highly specific policy language. Sending out a vague email saying, oh, feel free to use AI responsibly does not work. Right. Because what does responsibly even mean? You need to name the approved enterprise tools to find the specific use cases and set clear boundaries. The paper highlights a fantastic real-world example of this from Shopify. In April 2025, their CEO sent out a company-wide memo that was a masterclass in establishing legitimacy.

[00:39:22] Oh, Tobias Lütke's memo. Yeah. He didn't just suggest using AI. He made it a baseline expectation. He essentially told his entire workforce, you cannot opt out of this technological shift. Which is a bold move. It gets better. And furthermore, before you come to leadership asking to hire more headcount, you have to definitively prove that an AI agent couldn't accomplish the work first. It is an aggressive stance, undoubtedly.

[00:39:47] But it is highly effective because it leaves absolutely zero ambiguity about whether AI use is legitimate at Shopify. Right. There's no guessing. It is mandated. It flips the script entirely. Moderna took a similar, though perhaps more infrastructure-focused approach. What did they do? They legitimized the technology through broad, systemic deployment. They didn't just buy licenses. They built a custom internal chatbot that achieved 80% employee adoption. That's huge.

[00:40:14] And they actively deployed over 750 custom specialized GPTs across various departments. 750? When the tools are that widespread and heavily sanctioned, utilizing them becomes as normal as opening a spreadsheet. It becomes a standard part of the corporate furniture rather than a dark alley shortcut. Okay. So step one is making it undeniably legal. But just legalizing the tool doesn't automatically erase the lingering guilt or the fear of hallucinated mistakes. Right. The accountability piece.

[00:40:43] We have to change how we actually prove our work to each other. I love the framing the paper uses for this next shift. We need to replace the concept of confession with process transparency. Because a confession implies guilt. You only confess when you have committed a crime. That is a critical psychological distinction. The research clearly showed that the vulnerability employees feel stems from the act of disclosing something unusual or culturally muddy. Right.

[00:41:11] If you take the disclosure out of the realm of a whispered confession and turn it into a standard boring bureaucratic checklist, it completely loses its stigma. I love that. Make it boring. The goal for any organization must be to shift the managerial focus away from the binary question of whether AI was used and direct all the focus toward how the output was rigorously checked. The paper uses a fascinating example from the legal world to illustrate this.

[00:41:38] There is a federal judge in the Northern District of Texas, Judge Brantley Starr. Oh, yes. His standing order. Following the disastrous Avianca case, he didn't panic and ban AI in his courtroom. Instead, he issued a standing order that required all lawyers appearing before him to file a certification. A formal process. Right.

[00:41:57] They had to formally certify either that they did not use generative AI at all, or if they did use it, they had to certify that the AI-drafted language was meticulously checked for accuracy by a human being against traditional authoritative legal databases. That standing order is the gold standard for process transparency. It perfectly scales the burden of verification to match the level of risk. How so?

[00:42:22] Well, if you are an employee drafting a quick internal email about lunch plans, you don't need to fill out a rigorous process note. Sure. But if you are filing a brief in federal court, or to use another example from the paper, if you are a physician at Kaiser Permanente generating clinical medical notes with ambient AI scribes during patient visits, you need a documented, unshakable process of human review. Oh, that's a great example. In the Kaiser model, the patient gives explicit consent. The AI passively listens and drafts the notes.

[00:42:51] But the physician is required to explicitly review the text and take full human medical accountability for the final record. So they still own the final product. Exactly. There are no surprise exposures. There are no secret IT audits trying to catch people. There is just routine, expected process transparency. That makes total sense. It takes the emotion out of it, which leads right into the next major hurdle.

[00:43:17] If we are changing how we work, this fundamentally alters how we have to conduct performance reviews. The review process is entirely broken right now. We have to stop rewarding people just for looking tired and stressed out. We need to evaluate outcomes, not effort. And this requires a fundamental ground up rewiring of corporate performance criteria.

[00:43:37] If the perception of laziness is the core mediator driving this trust penalty, organizations have to forcibly direct their managers to look exclusively at the actual output. Right. We have to train leadership to ask different questions. Was the deliverable accurate? Was it delivered on or ahead of time? Did it drive measurable impact for the business? Not just did you sweat while making it? Exactly.

[00:44:01] And crucially, as we transition into this new era, we need to evaluate the judgment the employee demonstrated in producing it. But there is a massive, highly seductive danger here for HR departments, isn't there? The danger of relying on crude, easily trackable metrics. Oh, gamification. That's a huge risk. You can't just run a dashboard tracking how many prompts someone typed into the system and call that performance. The paper brings up the massive cultural backlash that happened at Duolingo.

[00:44:30] Yes, the Duolingo case study. In early 2025, their CEO put out a memo declaring a heavy prioritization of AI over human contractors. And they actually announced plans to officially consider AI use in employee performance reviews. And Duolingo serves as a vital cautionary tale of good intentions manifesting this terrible policy. Totally.

[00:44:52] If you communicate to your workforce, hey, we are going to track how many times you log into ChatGPT, and that number will impact your annual bonus, you are not measuring business outcomes. You are simply creating a new, bizarre, digital form of visible effort. People are just going to game the system. Employees are smart. They will optimize for the metric. They will sit at their desks, generating hundreds of useless, nonsensical prompts just to artificially inflate their numbers before review season. Just mashing the keyboard, basically. It doesn't drive productivity.

[00:45:21] It drives theater. Duolingo actually recognized this and had to formally reverse course a year later, clarifying to their staff that AI use no longer counted in reviews and wasn't expected if it didn't genuinely suit the task at hand. Contrast that approach with what companies like Meta and Microsoft are reportedly doing. They are explicitly tying the use of AI to performance impact. Impact being the key word. Right. They aren't measuring usage. They are measuring leverage.

[00:45:50] What specific business outcome did you achieve by utilizing this tool? It is a subtle semantic shift, but it is vital for healthy culture. Now, who is actually conducting these performance reviews? The managers. If the managers don't deeply understand the technology themselves, the new metrics are useless. We have to build manager AI fluency. Fluency is everything. And this might be my favorite insight in the whole paper because it relies on building empathy through actual hands-on exposure.

[00:46:19] The Reef study on the laziness penalty found a fascinating, highly actionable correlation. Managers who frequently use AI themselves do not penalize their employees for using it. Wow, really? Yes. When the evaluators in the experiment were regular AI users, meaning they used it weekly or more, they did not view AI-assisted employees as lazier. Because they actually know how it works. Familiarity breeds calibrated expectations.

[00:46:46] If a manager has personally spent three frustrating hours fighting with an LLM, trying to coax it into formatting a complex data table correctly, they know firsthand that using AI isn't a magical zero-effort wand. It still works. They understand that it requires specialized skill, iterative prompting, and rigorous verification. The paper highlights how Zapier tackled this. They didn't just tell managers to learn it.

[00:47:09] They ran company-wide interactive hackathons where everyone, regardless of technical background, had to build something using AI. That's a great way to do it. It was wildly successful. Their weekly active use jumped from 10% to over 50%. They made the learning process fun, communal, and hands-on. Exactly. But then you look at the complete opposite end of the spectrum with Coinbase. Let me ask you about this, because this feels highly relevant to a lot of intense corporate cultures. Oh, boy.

[00:47:37] In 2025, the CEO of Coinbase essentially issued an ultimatum to his engineering team. The message was roughly, adopt these new AI coding tools by the end of the week or you are fired. Very intense. And reports indicate they actually terminated individuals who didn't comply fast enough. I have to ask, is that Coinbase approach, just forcing adoption through the explicit threat of termination, actually building the kind of deep fluency and trust we need?

[00:48:03] Or is it merely replacing the employee's fear of looking lazy with a much more acute fear of losing their livelihood? It is almost certainly the latter. Coercion can absolutely drive rapid compliance in the short term, but it utterly destroys the foundation of psychological safety required for long-term innovation. Right. If I am using a complex new tool solely because I have a metaphorical gun to my head, I am going to use it in the safest, most superficial way possible. You're not going to take any risks. I am not going to push boundaries.

[00:48:32] I'm not going to experiment. I am certainly not going to raise my hand and admit when the tool generates a hallucinated error. Definitely not. And I'm absolutely not going to engage in the kind of honest, vulnerable dialogue required to establish real process transparency. You build genuine fluency through supported practice, peer-to-peer learning, and hands-on calibration. You do not build it through ultimatums and fear. Which perfectly sets up the final piece of the playbook.

[00:48:58] When mistakes inevitably happen, and they will, how do we handle accountability fairly? The fairness aspect is critical. If we know based on the data that older workers and female employees are taking the brunt of this unfair subconscious judgment, companies have a moral and operational imperative to audit who is actually bearing the penalty. The old management adage holds true here. You cannot manage what you do not measure. Right. Organizations must proactively monitor their AI adoption rates.

[00:49:28] And critically, they must break that data down by demographic groups. To see if there's a bias. They need to cross-reference adoption data with performance ratings and promotion rates. To see if individuals who transparently disclose AI use are systematically receiving lower compensation or fewer stretch assignments. The Butler-Snow law firm case from 2025 is a brilliant example of getting this accountability right in practice. Three lawyers at the firm were formally reprimanded for using ChatGPT and submitting fake citations to a court.

[00:49:57] Another fake citation case. But the internal sanction explicitly targeted their failure to verify the work. It did not punish the simple fact that they attempted to use a new tool to speed up their research. That is exactly what Dr. Westover refers to as accountability without stigma.

[00:50:13] You hold your professionals to the absolute highest, most rigorous standards of accuracy in client service, but you actively remove the moral judgment and the stigma surrounding the specific technological tools they utilized to try and achieve those standards. So, fixing immediate policies, mandating tools, updating rubrics, running hackathons, that is all vital. But how do organizations future-proof this mindset? How do we ensure we aren't just putting a band-aid on a symptom?

[00:50:42] You have to look at the long game. That is what the final section of our deep dive explores. The long-term horizon and the challenge of building what the academic literature calls calibrated trust. Calibrated trust is a profound concept. It's borrowed directly from the human automation interaction literature, tracing back to foundational work by Lee and See in 2004. Okay, what does it mean practically?

[00:51:06] It essentially means trusting a system, or in our context, trusting a human being who is using a system, to the exact precise degree that it deserves to be trusted based on its known capabilities and limitations. So balancing it out. We want organizations to avoid both dangerous extremes. We desperately want to avoid blind over-reliance, where overwhelmed managers just rubber stamp anything an AI touches because it looks confident and formatted nicely. Right.

[00:51:34] But we also must avoid stubborn underuse, where managers reflexively reject highly efficient, accurate work simply because of an outdated effort heuristic. You are aiming for the Goldilocks zone of trust. Not too hot, not too cold. Just the right amount of skepticism balanced with the right amount of faith. Exactly. And the absolute foundational prerequisite for achieving calibrated trust at scale is psychological safety. Right. The work of Harvard's Amy Edmondson from 1999 is seminal here.

[00:52:03] If a team environment is not safe for interpersonal risk-taking, you will never, ever achieve calibrated trust. Because people will just hide. An employee needs to feel fundamentally safe saying to their boss, hey, I used our LLM to draft this complex financial model. And frankly, here is a section where I think it might be hallucinating some of the projections. Can you sit with me for 10 minutes and help me check the logic? That requires so much vulnerability.

[00:52:26] If the employee fears they will be punished, demoted, or viewed as incompetent for making that admission, the entire system collapses back into the shadows. The shadow AI economy thrives in the absence of psychological safety. And this also means we have to fundamentally redesign how we bring people into our organizations in the first place. We have to move away from hiring and competency models that reward the ability to grind through manual labor.

[00:52:51] And also start writing job descriptions and designing interview loops that specifically screen for judgment. Right? Like how judgment is the new currency. We need to be asking candidates, did you pick the right tool for this specific problem? Did you know how to rigorously verify the output? And perhaps most importantly, do you possess the wisdom to know when not to use the AI? And finally, corporate governance cannot be a set it and forget it policy document. It has to become a continuous, real-time learning loop.

[00:53:21] How do they do that? Organizations need to deploy anonymous pulse surveys, similar to the methodology used by Slack and Charter in their research, to constantly measure internal concealment rates. Ah, track the hiding. Concealment is your canary in the coal mine. It is the ultimate leading indicator of organizational health in the AI era. If your pulse surveys show that your employees are still hiding their AI use, your team culture is profoundly sick.

[00:53:47] Regardless of how progressive or forward-thinking your official company policy claims to be. If you are listening to this right now, we want to bring all of this high-level theory directly back to your day-to-day reality. If you are a manager, a director, or a team lead, your action item for tomorrow morning is to talk openly with your team about an AI mistake that you made. Be the first to admit it.

[00:54:08] Show them with your own vulnerability that you are actively experimenting with these tools, that you are building verification into your own workflow, and that it is entirely safe to talk about the messy, imperfect reality of learning this technology. Break the ice for them. That's great advice. And if you are an individual contributor, your action item is to start utilizing process notes. Don't view it as a confession. Just routinely, professionally, and clinically document exactly how you verified your work.

[00:54:36] Prove your human accountability on paper. Make the extraordinary act of using AI completely boring. Make it boring is actually fantastic, highly pragmatic advice. True professionalism is very often about taking the novel, the chaotic, and the slightly scary, and operationalizing it into something routine, documented, and reliably dull. Okay, let's bring this all together. We started our journey visualizing two analysts on a high-stress Friday afternoon.

[00:55:05] And through that lens, we discovered a massive, deeply ingrained psychological blind spot that actively penalizes the very efficiency we claim to want. It's a huge paradox. We unpacked how this outdated effort heuristic is inadvertently driving a massive, global shadow AI economy, burning out our best employees, and creating staggering legal liabilities for companies who refuse to adapt. Right.

[00:55:29] And finally, we mapped out how clear, courageous leadership, routine process transparency, and a relentless managerial focus on verified outcomes are the only viable ways out of the trapdoor. But before we wrap up today's exploration, I want to turn it back over to you for one final thought. Out of all the research we poured through, what is the one concept that is really sticking with you? There is one particular finding from Shilke and Ryman's research on the transparency dilemma that keeps echoing in my mind, and it completely shifts the scale of the problem.

[00:56:00] What's that? They found that this disclosure penalty, this instinctual loss of trust, applies not just to individual employees talking to their managers, but to entire organizations communicating with the outside world. Wait, really? Their data show that established investment funds and creative graphic design firms demonstrably lost client trust the moment they openly disclosed they were utilizing AI in their workflows. Wow. The clients punish them.

[00:56:25] And if we connect this microbehavior to the macroeconomic picture, it raises a profoundly disturbing question for the future. Which is? If a company perfectly masters everything we just talked about, if they build flawless internal calibrated trust, if they foster total psychological safety and transparency among their own staff. But the outside world, their paying clients, their investors, their patients, still instinctively punishes them for it.

[00:56:51] Are companies ultimately destined to maintain strict organizational concealment from the public? Oh, man. Are we simply heading toward a macro-level global shadow AI economy where every corporation on Earth is highly automated, but we all just agree to pretend we are still doing things the hard way just to keep the consumers comfortable? That is an incredibly chilling thought to leave on.

[00:57:13] A whole world of hyper-efficient businesses secretly powered by algorithms, all putting on a coordinated theatrical performance of human struggle just to appease our outdated psychological biases. It's definitely something to think about. Thank you all for joining this deep dive into the hidden forces shaping the future of how we work, how we trust, and how we are judged. Think carefully about how you present your own processes tomorrow, stay curious about the systems you operate within, and we will catch you on the next one.