Amy C. Edmondson, 1999

“Team psychological safety is a shared belief held by members of a team that the team is safe for interpersonal risk taking.”

Amy C. Edmondson defined psychological safety this way in her 1999 study of learning behavior in work teams. Her research grew from a surprising observation in hospital data: teams reporting more errors sometimes appeared more effective, not less. The stronger explanation was that safer teams were more willing to report mistakes, making problems visible and learnable. Psychological safety concerns the interpersonal consequences of speaking up with a question, concern, idea, or admission. It helps explain why capable people may remain silent even when the organization formally invites candor and the stakes of withholding information are high.

For managers, safety is created through repeated responses. Leaders can frame work as uncertain and interdependent, acknowledge their own fallibility, ask specific questions, and respond appreciatively when someone raises a risk. The response need not agree with every suggestion. It must show that respectful contribution will be examined rather than punished. Teams also need accessible reporting channels, clear follow-up, and evidence that voiced problems lead to action. Safety becomes operational when information moves early enough to improve a product, prevent harm, or correct a decision.

Psychological safety is not comfort, consensus, or freedom from accountability. Edmondson distinguishes a learning zone in which high safety is paired with high standards. A group can debate rigorously, measure performance, and address poor conduct without humiliating people for uncertainty or honest error. Leaders who lower expectations in the name of safety create apathy, while leaders who demand excellence through fear create anxiety and concealment. The goal is candid, disciplined work in which people can surface what the team needs to know.

The concept is vital in distributed teams and AI-assisted workflows. Employees may notice hallucinations, bias, security gaps, or customer harm before formal metrics do, yet hesitate if enthusiasm for a launch feels mandatory. Teams can track near misses, questions raised, unresolved concerns, and the time from report to response. Retrospectives should examine the system, not hunt reflexively for a culprit. Edmondson’s definition offers a precise cultural test: when someone takes an interpersonal risk for the quality of the work, does the team make that risk worthwhile? Leaders can also rotate the first speaker in reviews and invite the person closest to the work before senior voices anchor the discussion. Written pre-reads and anonymous questions help when status differences are large. The meeting ends with an owner and response date for every material concern. These mechanics do not manufacture trust instantly, but they reduce predictable penalties for contributing information the hierarchy might otherwise miss.