AI Sentiment at Work Fell From 81% to 43%: What Boards Do Now
Two out of five employee reviews mentioning AI in mid-2026 were still positive. In 2019 it was four out of five.
AI Sentiment at Work Fell From 81% to 43%: What Boards Do Now
Two out of five employee reviews mentioning AI in mid-2026 were still positive. In 2019 it was four out of five. That collapse — 81% down to 43%, with 53% of AI mentions now openly negative — is the loudest workforce signal boards have ignored this year (Glassdoor analysis reported by The Decoder).
The split is not evenly distributed. Executives, software architects and managers still speak warmly about AI. Insurance claims adjusters, journalists, writers and frontline customer-service reps do not. Gen Z women recorded only 21% positive AI mentions — the most skeptical segment in the dataset. The complaint pattern is consistent: buggy tools pushed down by leadership, unclear accountability when the model errs, and productivity metrics that ignore the invisible cleanup work.
The governance gap the AI Act does not close
The EU AI Act’s high-risk provisions apply to systems, not to the deployment culture around them (Regulation (EU) 2024/1689). A HR-facing summarizer can be technically “low-risk” under Annex III and still produce the exact worker resentment that the Glassdoor numbers describe. ENISA has flagged the human-factor dimension in its 2024 threat landscape, noting that unmanaged AI deployments increase both insider risk and social-engineering exposure (ENISA Threat Landscape 2024).
NIST’s AI Risk Management Framework recognizes this: the Govern function requires organizations to track worker impact, feedback loops and role clarity — not only model metrics (NIST AI RMF 1.0). Most mid-market deployments we assess in our consulting practice have none of that in place. There is no logged channel for a claims adjuster to flag a summarizer that hallucinated a policy exclusion. There is no cadence for reviewing sentiment among the users. The tool ships, the metric dashboards go green, and the Glassdoor page absorbs the fallout eighteen months later.
worker_ai_governance:
feedback_channel: enabled
opt_out_paths: documented
cleanup_time_tracked: true
review_cadence_days: 30
escalation_owner: chief_of_staff
metrics_shared_with_workforce: trueA board-level response, not a comms exercise
When the C-suite is 40 points more optimistic than the people using the tools, the risk is not adoption. It is retention, litigation, and shadow-IT proliferation. The European Commission’s 2026 update on the Digital Decade targets ties workforce digital skills directly to competitiveness scoring for member states (EC Digital Decade Report).
Our working position at CAI Technology: the sentiment gap is a leading indicator of the same agentic-AI failures we track in production systems — misaligned incentives at the top, silent degradation at the edge. Fix the deployment culture first, then the tools. If your board wants to see where your organization sits on that curve, our consulting team runs a two-week diagnostic on worker-AI trust and governance readiness.