Employees Hide AI Workflows Due to Trust Gaps, Not Policy

Employees Hide AI Workflows Due to Trust Gaps, Not Policy

A physician developed a prompting template for his organization's approved AI tool that "produces astoundingly good results." His colleagues struggled with the same tool and told him so. He knew his template would help them.

He never shared it.

This pattern is playing out across organizations worldwide. While companies respond to hidden AI use with more governance—usage policies, approved tool lists, monitoring systems—new research suggests they may be solving the wrong problem. The real barrier to knowledge sharing isn't inadequate rules. It's inadequate trust.

Why it matters

Unlike past productivity gains embedded in shared systems, AI innovations often emerge from individual experimentation—a prompt sequence that cuts a three-hour task to 20 minutes, a workaround that bypasses a process bottleneck. When employees keep these discoveries private, organizations lose compounding productivity gains that could spread across teams. The suppression isn't about problems or risks; it's about solutions that never scale.

Trust predicts sharing more than policy

Researchers Eric Anicich and Jeslyn Brouwers surveyed 604 U.S.-based employees who use AI daily or multiple times per day, as first reported in Harvard Business Review. Nearly one in three (30.3%) admitted intentionally withholding AI-related knowledge, workflows, or techniques from coworkers or employers.

The strongest predictor wasn't whether the organization had an AI policy or provided approved tools. It was organizational trust. Employees in the lowest trust quartile were nearly four times as likely to withhold AI knowledge as those in the highest quartile—47% versus 14%. A similar gap appeared for psychological safety (45% versus 17%).

Neither having an AI policy nor access to approved tools predicted knowledge sharing on its own. Trust remained a strong predictor even after accounting for job insecurity, workplace competition, perceptions of fairness, age, gender, industry, and job level.

The relationship between trust and knowledge-hiding weakened considerably after accounting for psychological safety, suggesting trust creates the willingness to share while psychological safety provides the environment where sharing feels safe.

Three costs of visibility

Interviews revealed employees make rational calculations about disclosure costs. First is reputational risk—a junior consultant noted colleagues avoid discussing AI use because it makes them "seem less capable." An analyst described a colleague who shared an AI note-taking feature only to be discredited by a senior team member who dismissed work done "by a computer."

Second is workload expansion. A management consultant explained: "If I automate A and B, they're not just gonna let me focus on C. They're gonna make me do D, E, F." When efficiency gains are rewarded with more work rather than better work, employees keep methods private.

Third is replaceability. Enterprise AI systems can record prompts and document workflows, creating detailed maps of employee methods that can be transferred to replacements or automated entirely. When hiding from your employer becomes prudent career strategy, the trust battle is already lost.

What leaders should do

Organizations need clear guidance on encouraged versus prohibited AI use, removing ambiguity that forces employees to manage appearances. Use lightweight templates and "show me how you built this" sessions to convert private methods into reusable artifacts. Structured conversations about techniques work—one field experiment with salespeople found they produced average sales gains exceeding 15% that lasted at least 20 weeks.

Stop taxing efficiency gains. Leaders need explicit norms for how saved time will be used—deeper analysis, higher-value projects, professional development—so employees see upside, not just extraction.

Reward multiplier behavior, not just individual productivity. Give credit in performance reviews for methods others adopt, protected time to experiment, and a share of gains once workflows spread. Team-based compensation generates more knowledge transfer than individual commissions, research shows, but only when combined with pro-sharing norms.

These findings were first reported by Eric Anicich and Jeslyn Brouwers in Harvard Business Review, based on their survey of 604 daily AI users and interviews with professionals from analysts to CEOs.