The Disclosure Paradox
- Wickersham Team
- 14 minutes ago
- 7 min read
Transparency Can Change the Work Before Anyone Sees the Label
A new study found that telling creators their AI use would be disclosed changed how they created, before the audience ever saw anything. The finding extends far beyond AI. What an organization expects to explain later shapes how its people behave now.
The study was not primarily about how audiences respond to AI disclosure. That has been studied elsewhere, with reasonably predictable results: audiences tend to rate AI-assisted work less favorably when they know about the AI involvement, a bias that holds even when the AI-assisted work is objectively better.
What this study found was different, and considerably more useful. A paper published in Information Systems Research in August 2026 examined what happened when creators knew in advance that their use of AI would be disclosed to the audience. The result was not primarily a finding about audience perception. It was a finding about creator behavior.
Many creators who anticipated disclosure withdrew more from the creative process. They allowed the AI to do more of the work. They contributed less of their own judgment, revision and creative agency. The work became more computational and less human, not because the creators were lazy or indifferent, but because they anticipated that the audience would discount their creative contribution regardless. The disclosure policy designed to protect human authorship inadvertently suppressed it.
That mechanism is the finding worth building a framework around. Not what audiences do with disclosure, but what creators do with the anticipation of it. The distinction matters because it shifts the entire conversation from how to communicate transparency to what transparency actually produces in the people it is designed to govern.
The Mechanism Behind the Paradox
The logic of the creator's response is not irrational. It follows directly from what they expected to happen.
If the audience discounts my contribution because AI was involved, then it will be discounted regardless of how much I actually contributed. The effort I invest in revision, in judgment, in the distinctly human parts of the work, will not be credited. The label will characterize the output as "AI," not the degree of my involvement. Given that, why invest the effort that will not be recognized?
This is the behavioral trap that anticipated disclosure creates. It produces a rational withdrawal from the very thing the disclosure policy was meant to protect. The policy creates the condition it was designed to prevent.
The finding has a specific implication for how organizations think about governance mechanisms generally. A disclosure requirement is a governance mechanism. So is a legal review process. So is a leadership approval requirement. So is a client transparency policy. So is an audit trail. Each of these is a form of anticipated accountability, and each can change behavior among the people subject to it before any external party ever evaluates the output.
Governance mechanisms do not only communicate to the audience. They communicate to the people being governed. What they communicate upstream, before any audience arrives, is qhere the most consequential behavrioal effects live.
The Organizational Patterns Worth Examining
The AI disclosure finding is the clearest version of the paradox because the mechanism is explicit and the study measured it directly. But the same dynamic operates in organizational contexts unrelated to AI, in ways most leaders have never examined.
Consider what happens when a writer knows that legal will review the final copy. The first draft written is not the writer's most interesting. It is the draft that anticipates the legal review and preemptively removes anything that might create friction during that review. The result is copy a lawyer has never challenged, but that carries the conservatism of anticipated legal challenge in every sentence. The writer did not become less capable. They became more defensive, for rational reasons, in response to a governance mechanism that was not yet active.
Consider what happens to a strategist who knows that a recommendation must go to leadership for approval before it reaches the client. The recommendation that gets developed is not the strategist's most provocative thinking. It is the version that anticipates the leadership conversation and calibrates to what is likely to survive it. Genuine strategic insight that requires a difficult internal conversation may never get written, because the person who would write it already knows it will not pass the gate. The governance mechanism has shaped thinking before it is complete.
Consider what happens to a designer who knows the client will ask whether AI was involved in producing any element of the work. The design process that follows may avoid AI tools that would genuinely improve the outcome, not because the designer was instructed to avoid them, but because the anticipated disclosure conversation creates a friction cost that makes avoidance feel easier than explanation. The governance mechanism has changed the process without ever being applied.
In each case, the mechanism is the same: the person doing the work is adjusting their behavior in response to anticipated accountability, and that adjustment produces a different output than would have been produced without the anticipation. The adjustment may be rational from the individual's perspective. It is rarely the output the organization intended to receive.
Why This Is a Leadership Problem, Not a Compliance Problem
Most organizations treat disclosure, approval, and review requirements as questions about process and accountability. Who signs off? What gets documented? What does the audience see? These are legitimate questions, and they deserve answers.
What they are not asking is: what does this governance mechanism communicate to the person whose work it governs, before anyone else is involved? And is the behavior that communication produces the behavior the organization actually wants?
These are different questions. The first set is about the output end of the process. The second set is about the input end. And the ISR study suggests, with some precision, that the input end is where governance mechanisms do their most consequential work, including the work that organizational leaders have not sanctioned and may not be aware of.
A legal review process that produces self-censored first drafts is not only a legal process. It is a creative brief. A leadership approval requirement that produces recommendations calibrated to what will survive the approval meeting is not only a governance mechanism. It is a filter on strategic ambition. A client transparency policy that leads to the avoidance of tools that would improve the work is not only unethical. It is a design constraint.
Leaders who have not asked what their governance mechanisms are communicating upstream, to the people doing the work, are receiving outputs shaped by those communications without knowing it. The gap between the work they think they are receiving and the work that is actually being produced is the organizational cost of the Disclosure Paradox.
A governance mechanism shapes the work before it governs it. Most organizations are measuring what the mechanism produces at the end. Almost none of them are measuring what it changes at the beginning.
What Transparency Actually Is
The conclusion this article does not make is that transparency is bad, or that governance mechanisms should be removed. That would be the wrong overcorrection, and it would miss the more interesting insight the research provides.
Transparency is genuinely valuable. Disclosure of AI involvement, legal review of consequential communications, leadership oversight of strategic direction: these serve real organizational purposes, and they should. The research is not an argument against them.
The research argues against the assumption that transparency is neutral. That implementing a disclosure requirement, an approval process, or a review mechanism produces only the intended effects on the audience side, while leaving the behavior of the people being governed unchanged. That assumption is not supported by the evidence.
Transparency is a behavioral environment. The people operating inside it are responding to its signals constantly, before the audience ever arrives. When those upstream behavioral responses are not examined, organizations end up with governance mechanisms that work as intended at the output end but produce unintended consequences at the input end.
The more sophisticated version of transparency asks both questions simultaneously: what does this mechanism communicate to the audience? And what does it communicate to the people doing the work? Designing governance mechanisms that serve the first purpose without undermining the second is harder than implementing a disclosure policy. It is also the work that actually produces the outcome the policy was designed to achieve.
The Diagnostic Worth Running
For any governance mechanism currently in operation, a useful audit examines upstream behavior rather than downstream compliance.
What do people do differently when they know this mechanism is coming?
Does the anticipation of legal review produce more careful drafting, or more defensive drafting?
Does the anticipation of leadership approval produce more ambitious recommendations, or more calibrated ones?
Does the anticipation of client transparency requirements produce more thoughtful process choices, or more avoidance of tools that would require explanation?
These questions do not have universal answers. The behavioral effect of anticipated governance depends on the specific mechanism, organizational culture, and people subject to it. But the questions are worth asking, because most organizations have never asked them, and the answers would change how governance mechanisms are designed.
The ISR study found that anticipated AI disclosure made human creative contributions less visible by reducing human creators’ willingness to contribute. A governance mechanism intended to surface human authorship was reducing it. That is the paradox. And it is not unique to AI.
What an organization expects to explain later changes what its people do now. That dynamic is not a malfunction of the governance mechanism. It is a feature of how people operate within accountability systems, responding to anticipated judgment before it arrives.
Organizations that design governance mechanisms without examining that dynamic are not implementing transparent processes. They are implementing processes whose most significant effects are happening in the part of the system no one is watching.
The question worth asking of every disclosure requirement, every approval process, every audit trail, is not only what it communicates to the audience at the end. It is what it communicates to the people at the beginning. That is where the work actually gets made, or unmade, before anyone sees the label.
Some ideas are worth discussing in the context of your organization.
