IBM study finds workers worry AI erodes judgment as review work grows

A global IBM study released September 21 finds a striking gap between what human resources chiefs say workers need from AI and what workers say matters to them. In the survey, 71% of chief human resources officers named the ability to supervise, validate and override AI outputs as an essential skill. Only 29% of employees ranked judgment as important. IBM's message is that using AI at work is becoming a management and accountability problem, not just a software rollout.

Workers see a cost behind the efficiency story

IBM and Oxford Economics surveyed 1,500 CHROs and 8,800 full-time employees between April and June 2026. The employee sample spans 28 countries, while the executive sample covers 21 geographies and 23 industries. These are reported perceptions from different groups, not a controlled measure of how individual skills changed.

Still, the concerns are concrete. Sixty percent of employees said they worried AI was eroding their skills, most often pointing to critical thinking. Among workers with that concern, three in four said some erosion had already begun. Forty-three percent reported that when something goes wrong with AI, blame falls on them. Forty-two percent said AI increased their work or that the extra work went unrecognized.

That extra labor can hide inside a productivity metric. Eight in ten CHROs said AI adoption creates invisible tasks such as checking recommendations, fixing errors and handling exceptions. A team may automate a first draft while adding more verification work downstream.

Responsibility is still fragmented

IBM found that 46% of organizations did not involve their CHRO when defining AI strategy. Only 28% of CHROs reported a shared roadmap and regular operating cadence with IT. The study also says 41% of CHROs believe employees may not feel safe challenging or overriding AI outputs.

A notable association runs the other way. Where CHROs share responsibility for deciding which choices remain human-led, 76% of employees said they felt safe questioning or overriding AI recommendations, compared with 43% where HR is only advisory. The survey cannot prove that the governance arrangement caused that gap. It does, however, show why an instruction to keep a human in the loop is weak unless that human has authority and time to act.

The study's bigger test

IBM says organizations that explicitly sort workflows into human-led, AI-assisted and AI-executed categories report lower risk and better quality. Those performance comparisons rely on the study's own analysis and self-reported business outcomes, so they should be tested against independent measures before being treated as universal benefits.

For employers, the immediate question is who owns the review burden when AI produces a plausible but wrong answer. Workers need clear escalation paths, protected time to check results and a way to retain judgment as routine tasks move to software. Otherwise, faster output may simply make mistakes and responsibility harder to see.