What the Manager Cannot See
A manager in a large engineering services firm sat across from me last month, frustrated. Her organization had invested in a rollout of AI productivity tools nine months earlier. The dashboards said adoption was strong. The throughput metrics were trending in the right direction. Her team's engagement scores, however, had quietly slipped by twelve points over the same period, and she did not know why.
I asked her one question: Do you think your team is using AI more, or less, than the dashboards say?
She thought for a long moment and said: "More. They are using it more. They just are not telling me."
This is now, in my consulting practice, one of the most consistent patterns of 2026 — and one of the least well-named. The team is using AI. The manager suspects they are using AI. The team suspects the manager suspects. Nobody is talking about it. And the silence is doing slow, expensive damage to the relationship that the manager's job actually depends on.
The Finding That Explains Everything
A series of ten experiments conducted by researchers at the Wharton School and UC San Diego, totalling 3,346 participants, asked managers to allocate bonuses to workers who had produced identical, quality-verified output. The only difference between the workers was whether they had used AI assistance to produce it.
Managers reduced the bonuses by approximately 50% for the AI-assisted workers — roughly $0.35 per unit versus $0.65 for the workers who had produced the same output unaided. The reduction held even when the manager could plainly see that the human's contribution had been substantive.
The mechanism the researchers identified is straightforward and uncomfortable: attribution drift. When a worker uses an AI, the manager unconsciously credits the AI for the output and reduces the worker's perceived deservingness accordingly. The worker who quietly polished the AI's draft is judged to have done less than the worker who hand-typed an inferior version. The judgement is not based on the quality of the result. It is based on the manager's narrative about who did the work.
Your team has noticed this. They may not have read the Wharton paper, but they have watched a colleague's contribution get described as "mostly the AI" in a calibration meeting. They have seen a peer's promotion case weakened because the work was visibly AI-assisted. They have, almost certainly, drawn the rational conclusion that the safest behaviour is concealment.
What Concealment Looks Like
Concealment, in this context, is not deception. It is risk management. The patterns I see most often:
The Polish-and-Submit pattern. The worker uses the AI to generate a draft, spends real time refining and verifying it, and presents the final product without mentioning the AI's role. When asked, they say something accurate but incomplete: "I worked on it over the weekend."
The Pre-Drafted Conversation. The worker walks into a meeting having privately run their position through an AI, anticipated counter-arguments, and stress-tested their framing. They present as confident and prepared. They are confident and prepared. They are also operating with a level of preparation that the manager cannot price in to their assessment of the worker's underlying capability.
The Selective Attribution. The worker openly attributes the boring work to the AI ("I had the AI summarize the meeting notes") while quietly absorbing the credit for the interesting work that the AI also helped with. Over time, the manager's mental model of the worker's contribution drifts further and further from reality.
The Audit-Adversary Drift. Where the work is recorded and visible — code reviews, project management tools, internal documentation — the worker's behaviour optimizes for what the recorded trail will look like rather than what the work actually requires. They reject some AI suggestions not because the suggestions are wrong but because rejecting them produces a defensible story of independent judgment. The recorded trail looks excellent. The thinking behind it has shifted.
None of these patterns are signs of bad employees. They are signs of intelligent employees responding to a reward system that has not yet caught up to the technology. The patterns are produced by the system; they are not character flaws.
The Cost the Dashboard Will Not Show
The immediate cost of concealment is informational. The manager loses the ability to develop the worker because they cannot see what the worker is actually doing. Coaching conversations become hollow. Calibration meetings become guesswork. Promotion cases get built on partial evidence.
The deeper cost is relational. A worker who is concealing the conditions of their work cannot be in a fully candid relationship with their manager. The relationship that the manager's role most depends on — psychological safety in both directions — quietly hollows out. The worker stops bringing the manager problems early because bringing problems requires explaining how the work is being done. The manager stops getting the kind of unguarded signal that lets them catch issues before they become incidents.
I see the downstream consequence in retention data. The OECD's 2025 study of approximately 6,800 workers across seven countries found that algorithmic management is associated with reduced autonomy, lower trust, higher stress, and reduced job satisfaction — except in workplaces where employees had substantial influence over how the systems were configured and used. In those workplaces, the negative effects did not merely soften; they disappeared. Worker voice was not one variable among many. It was the variable.
If your team is concealing their AI use from you, you have, by definition, the low-voice version of this situation. The outcomes will follow.
Why This Is a Middle-Management Problem Specifically
Senior leaders are mostly insulated from this dynamic because they do not personally evaluate the work that AI is now helping to produce. They evaluate metrics. The metrics look fine.
Individual contributors are not the ones holding the bag, either. They are simply adapting.
The middle manager is the joint where the friction concentrates. They are evaluating work whose production conditions they cannot see, on behalf of leadership who assume the metrics tell the full story, with a team that has correctly identified disclosure as a career risk. They are squeezed from above and below, with no formal training in how to hold the squeeze.
This is the structural reason that middle management engagement is, in our 2026 client data, the steepest negative trend line in organizations one year into significant AI adoption. The issue is rarely the technology. The issue is that the manager is the only person in the chain whose role requires them to bridge the gap between what is recorded and what is actually happening, and most managers were never developed in the emotional capacities that bridge demands.
What the Bridge Actually Requires
The capacities that distinguish managers who navigate this transition well from those who do not are, in my coaching practice, remarkably consistent. They are not technology capacities. They are emotional intelligence capacities exercised in a new context.
- Attribution discipline. The ability to credit the human contribution accurately when an AI was involved — not generously, not punitively, but accurately. This is harder than it sounds, and it has to be practised against the manager's own attribution drift, not just modelled for the team.
- Disclosure-safe inquiry. The ability to ask how a piece of work was produced in a way that does not punish honest answers. Most managers, asked to do this, will produce a question whose phrasing already contains the threat. The team hears it.
- Calibrated trust. The ability to hold the difference between "I trust this person" and "I trust the output this person produced under conditions I cannot fully see." These are different epistemic positions and they require different managerial responses.
- Reward-system advocacy. The ability to push upward against incentives that are producing concealment in the team. Most middle managers do not see this as part of their job. It is now the largest single lever they have over team behaviour.
- Relational repair. The ability to recognize when concealment has already taken root — and to reopen the conversation without re-creating the conditions that produced the silence in the first place.
These are coachable. None of them are intuitive. All of them are unevenly distributed across the manager population of any large organization, and the unevenness now matters in ways it did not matter five years ago.
Where to Look First
If you suspect this dynamic is operating in your team — and the base rate, in our client work this year, is that it is — the first step is not a policy change. It is a diagnostic. Policy changes that get rolled out without first understanding the underlying pattern of concealment tend to deepen it; the team reads the policy as evidence that the manager is closing in, and adapts accordingly.
Our EQ Pulse organizational health diagnostic is designed for exactly this question. It measures the gap between the team's surface engagement signals and what is actually happening underneath them — including the dimensions that the Wharton finding predicts will be most affected by attribution drift. It is a quiet instrument, designed to be safe for the team to engage with honestly. That last property is the one that determines whether the diagnostic produces anything useful at all.
For managers specifically, we have written more about the structural side of the middle-manager squeeze on our solutions for teams page — what the role looks like in an AI-augmented organization, why the conventional middle-management playbook is producing the opposite of its intended effect, and what coaching can do to rebuild the bridge the role now requires.
The manager I spoke with at the start of this piece was right. Her team was using AI more than the dashboards said. The work of the next six months is not to catch them at it. It is to build the conditions under which there is nothing to catch — because the team is no longer paying a price for telling the truth.
If you suspect concealment is operating in your team, start with the EQ Pulse diagnostic, or read more about how we work with middle management on our solutions for teams page. For the broader context, see our pillar piece on the AI Anxiety Stack — where the manager's attribution fear sits inside a three-layer model with engineer and leader fears.