Organizations are asking employees to adopt AI and they are also asking them to disclose when they use it. Both expectations make sense. Adoption supports productivity while disclosure supports responsible AI and accountability. But emerging research reveals a tension inside them. Employees can be encouraged to use AI and evaluated less favorably when others know they used it. That is the trap. Neither expectation creates it alone. It forms between them, inside an evaluation system many organizations never examined.

Adoption creates a second threshold

A 2025 working paper examined a generative AI coding assistant rollout at a large technology company. Digital records from 28,698 software engineers showed uneven adoption despite company incentives. Twelve months after the rollout, 41% of engineers overall had adopted the tool, compared with 31% of women.

In a preregistered experiment, engineers rated identical work differently depending on whether they believed AI was used. The reported competence penalty was 9% overall, rising to 13% for women compared with 6% for men. In addition, survey data showed an association between anticipated competence penalties and slower adoption. This is one company, one occupation, and a working paper. It does not establish how professionals in other industries will respond.

Once professionals begin using AI, they encounter a second threshold: whether they can disclose that use without changing how their competence is evaluated. An adoption dashboard records participation. It cannot show whether professionals feel safe enough to be transparent about how that participation occurred.

The social cost is not limited to software engineering

In four preregistered experiments with over 4,400 participants, workplace AI users anticipated and received more negative evaluations of their competence and motivation. Those judgments also affected evaluations of job candidates. Evaluators who used AI themselves were more willing to hire AI users, so manager experience appears to influence individual decisions. But one evaluator's experience is not a shared norm and the norm determines how disclosure gets read.

In a separate study published in Organizational Behavior and Human Decision Processes, 13 experiments found that disclosing AI use reduced trust in the user. Reduced perceptions of legitimacy helped explain the effect. A favorable view of technology softened the penalty but did not remove it.

Teaching more people to use AI may not, by itself, change what AI use signals inside the organization. These experiments generally involved participants evaluating strangers, not managers reviewing employees they know. The effects may not transfer unchanged into every workplace. Still, AI adoption has social conditions, not only technical requirements.

Concealment is not always a governance failure

Organizations frequently discuss undisclosed AI use as shadow AI, a governance and compliance problem. Those risks are real. When a workplace requires transparency while its evaluation system attaches a professional cost to AI use, concealment becomes predictable. The employee may not be rejecting AI. She may be using it well while remaining uncertain about what disclosure will cost her.

Will my manager believe the AI did the important part?

Will efficiency be interpreted as reduced effort?

Will my judgment receive less weight?

This distinction matters for Quiet Technophobia®.

Quiet Technophobia® names hesitation beneath competence. Concealment after adoption is a different behavior and the two should not be collapsed into one construct. But they may be shaped by the same evaluative environment. Hesitation before use; concealment after use. One set of workplace conditions may contribute to both. This is not a diagnosis of either professional. It is a reason to examine the environment surrounding adoption.

AI governance must prepare the evaluator

AI governance often starts with approved tools, disclosure requirements, data protections, and human-in-the-loop standards. A disclosure policy tells employees what they must reveal. It does not prepare managers to interpret that disclosure fairly. Leaders should examine two things before treating low adoption or hidden use as an employee deficit:

The answer is not less disclosure

Third-party exposure of undisclosed AI use damages trust more than voluntary disclosure. Responsible AI still requires appropriate transparency and accountability. A generic disclosure statement may not be enough. The 13-experiment trust study tested six framings, including that a human had reviewed the work and that AI was used only for proofreading. Every framing was trusted less than no disclosure.

Research in professional services suggests a more useful direction. Across six experiments with 1,664 participants in healthcare, legal consulting, and business consulting, disclosure emphasizing complementary capabilities and the professional's choice to use AI (self-selection) reduced the negative effect. That shifts disclosure from:

"AI was used."

to:

"Here is the role AI performed, the role the professional retained, and who remained accountable for the result."

That gives evaluators a more accurate basis for judgment. Organizations cannot measure meaningful AI adoption without examining what happens after disclosure. Participation is only one threshold. Transparent participation is another. If transparency carries a professional penalty, quiet AI adoption should not surprise us.

The evaluation problem underneath this one has a name.

When identical work is judged differently depending on how it was produced, the organization is measuring method, not result. That is Process Purity operating inside an evaluation system and it is one expression of Caste Coding.

I wrote about how that gatekeeping produces the hesitation it claims to solve: UNIVITED

To learn more, visit my website.