The silent cost of AI at work: how AI erodes psychological safety

A 2025 peer-reviewed study in Humanities and Social Sciences Communications, part of the Nature portfolio, directly measured what happens to teams as AI adoption rises. Kim, Kim and Lee studied 381 employees across three time points and found that AI adoption showed a statistically significant negative association with psychological safety: as AI use increased, employees reported feeling markedly less safe speaking up, admitting errors, or taking interpersonal risks. More troubling, psychological safety mediated the relationship between AI adoption and employee depression. AI didn't just make people quieter, it made them unwell, and the erosion of safety was the pathway. This isn't a theoretical concern, it's a measurable, documented organisational risk.

Three mechanisms explain why AI makes people go quiet. Job insecurity and identity threat: when AI touches the thinking, analysing, and deciding that professionals built careers around, admitting confusion starts to feel like evidence of becoming replaceable, so people stop asking. Loss of autonomy: when decision-making shifts to an algorithm, people become less willing to question it, because if 'the system' made the call, raising a hand to say you're unsure feels riskier, not safer. And chronic uncertainty and cognitive overload: constant change to roles, skills, and workflows depletes psychological resources and pushes people into defensive, low-risk behaviour. The scale of that exhaustion shows up in workforce data: a McKinsey Health Institute survey of over 30,000 employees across 30 countries found one in five professionals experiencing burnout symptoms tied partly to rapid AI-driven role change, while ManpowerGroup's 2026 data found AI workplace usage up 13% year over year and employee confidence in the technology down 18% over the same period. More tools, less confidence, isn't a team ready to take creative risks, it's a team in survival mode.

Amy Edmondson's more recent work with Jayshree Seth, writing in Harvard Business Review, adds a layer she calls 'trust ambiguity.' When a human colleague makes an error, teams typically rally around it, discuss it, and move on. When AI makes an error, the response is often diffuse, directionless doubt: people aren't sure what went wrong, whether it will recur, or whose job it was to catch it. That ambiguity inhibits exactly the behaviours that define psychological safety: speaking up, questioning, and learning.

The research points toward specific interventions, and the most important variable in all of them is leadership, not technology. Making AI's uncertainties explicitly discussable, asking directly 'how is AI affecting your collaboration' or 'where are you uncertain about AI's role in your decisions', surfaces trust issues before they calcify into silence. Silence isn't agreement, it's often the first symptom of the Silent Resistance I describe elsewhere. Treating an AI rollout as a team-development challenge rather than a pure technology deployment matters just as much: involving teams in decisions about how tools get used, creating feedback loops, and acknowledging openly that the integration will be messy. Kim, Kim and Lee found one critical moderator across all of this: ethical, transparent leadership significantly buffers the negative impact of AI adoption on psychological safety, and for the highest-stakes conversations, face-to-face interaction matters more, not less, since remote and hybrid formats already suppress candour before AI-related anxiety gets layered on top.

Most organisations track AI adoption rates, efficiency gains, and cost savings. Almost none systematically measure psychological safety, employee voice, or willingness to report AI-related errors, which is exactly the gap the Organisational Adoption Profile is designed to close. Before your next AI initiative, ask your team one question: what's the worst thing that could happen if you tried this and it didn't work? If you hear 'I'd learn something,' you have psychological safety. If you hear silence, nervous laughter, or 'I'd get in trouble,' you have your diagnosis.