Psychological safety is not what one person feels. It's what a team, under a leader, produces together. That distinction — relational rather than individual — is the argument at the heart of Safe to Great (2023), and it changes how you build psychological safety in practice. This guide gives you Amy Edmondson's foundation, the six-stage developmental model that operationalises it, and the bridge into growth mindset that most organisations never cross.
Psychological safety is the shared belief within a team that interpersonal risk-taking is safe. Amy Edmondson coined the term at Harvard in the 1990s, working with hospital teams. She noticed something odd: the teams that reported the most errors were the ones with the best performance. Not because they made more mistakes — because they were willing to talk about them.
That's the foundation. Psychological safety means people can raise concerns, admit errors, challenge ideas, and propose new approaches without fear of humiliation, retaliation, or social cost. Google's five-year Project Aristotle study confirmed it as the single most consistent predictor of team performance — above technical skill, above seniority, above any other variable they tested. Barbara Fredrickson's research at UNC adds the emotional science: safety produces positive emotions (trust, curiosity, confidence) which in turn broaden people's creative and problem-solving capacities.
So far, so straightforward. This is the part most articles about psychological safety stop at. It's also the part most organisations claim to have. They rarely do.
Psychological safety is not a feeling. It's a condition teams produce together, under leaders who make it possible.
Here's the uncomfortable bit. Human DNA doesn't favour psychologically safe teams. It favours protective teams — tight, loyal, single-identity groups built for survival. That's evolutionarily sensible: predators, scarcity, out-group threat. Protective teams feel good because they feel safe, and they feel safe because they screen out difference.
The problem is that protective teams are terrible at learning. They punish dissent, suppress diverse views, and treat challenge as disloyalty. They feel cohesive right up until they're overtaken by a team that actually learns.
Psychological safety is a cultural achievement, not a biological default. It runs against our wiring. Which is why it doesn't happen by accident, doesn't survive leadership change, and doesn't hold under pressure unless the system is explicitly built to hold it. Every workplace drifts back toward protection when nobody's watching. The question for leaders is whether they notice the drift.
Edmondson's real breakthrough was not defining psychological safety. It was establishing that it's interpersonal, not individual. Before her research, organisational performance was overwhelmingly explained through individual variables: talent, competence, motivation, personality. She demonstrated that the quality of relationships within a team was a stronger predictor of learning and error reporting than the skill of any single member.
That relational insight is the part worth defending. It's also the part most of the later conversation quietly walked away from.
Somewhere between the academic research and the HR training deck, the concept drifted. "Team psychological safety" became "individual courage to speak up." The leader's accountability for creating the conditions got replaced by the employee's responsibility to be brave. The word safety started doing more work than it could carry — sometimes meaning comfort, sometimes meaning emotional validation, sometimes meaning "don't give me difficult feedback."
That drift was the subject of a piece I published on Medium in February 2024 titled "What Amy Edmondson Gets Wrong About Psychological Safety." The title was deliberately provocative. The argument was not: my bone to pick was never with the research, but with how we apply it. Edmondson herself engaged with the point publicly on LinkedIn — agreeing, in substance, that the relational core of the concept had been thinned out in its popular translation.
In May 2025, Edmondson and Michaela Kerrissey published "What People Get Wrong About Psychological Safety" in Harvard Business Review — an article cataloguing six misconceptions about the concept she originated. The arc matters: by 2025, the Edmondson–Kerrissey piece in HBR was making the case that the popular understanding of psychological safety had drifted from the research. Safe to Great was published in 2023, and had been making a related argument from the relational angle since then. I'm not claiming originality over the founder of the field. I'm noting that the relational correction was already in the book, and that the conversation has since moved in that direction.
The popularity of the term psychological safety has, in many cases, undermined the important insights of the theory.
Here's the positioning that Safe to Great anchors. Psychological safety is relational. It is produced between people — by leaders, through teams, across hierarchies — not generated inside individual heads. A person cannot be psychologically safe alone. The condition only exists in relation.
This matters because it relocates accountability. If safety is individual, the fix is courage training — teach people to speak up. If safety is relational, the fix is leadership behaviour — hold leaders accountable for the environment they create. These are wildly different interventions, and they produce wildly different results. Courage training without environmental change produces the same silence plus guilt.
Relational psychological safety also has a twin concept: relational growth mindset. Both terms appear in Safe to Great as a deliberate correction to the individualistic drift in both fields. Carol Dweck's original growth mindset research focused on individual belief. Amy Edmondson's original psychological safety research focused on interpersonal team conditions. Popular application collapsed both into individual mental work. Safe to Great re-establishes them as relational — and as two parts of the same continuum.
To hold the relational view without dismissing the individual experience, you need two categories. Safe2Great distinguishes between inner and outer psychological safety — and both matter.
Outer safety is the actual environment. It's relational and based on what really happens. When I say something that matters to me, do I get listened to? Do I get criticised? Do I get ignored? Do I get punished later? Fears about outer safety are not imaginary — they are rational responses to actual negative consequences. Your boss really can be an arsehole, and sometimes is.
Inner safety is the individual's internal capacity to tolerate the discomfort of risk-taking without retreating into a protective mindset. It's the work of noticing your own fight, flight, or freeze responses and choosing not to be run by them. In Safe2Great language, the three protective mindsets are the Hippo (controlling-competitive, the fight response), the Snail (complying-complacent, the flight response), and the Clam (critical-sceptical, the freeze response). Each offers a false sense of security that feels protective in the moment and undermines everything over time.
The interplay matters. Outer safety without inner safety produces dependence — people who need a perfect environment to function. Inner safety without outer safety produces isolation — individuals regulating well in genuinely unsafe conditions, often burning out as they do. Real psychological safety is both: a team environment that rewards honesty, and individuals capable of bringing honesty to it.
If psychological safety is relational, you need a model that shows where a team is and what comes next. The Safe2Great six-stage model does that. Each stage names a different developmental capacity. Teams move through them in order. Skip a foundational stage and the higher stages don't hold.
Unconditional acceptance. People feel connected and valued irrespective of role, location, background, or current performance. This is the bedrock. Without belonging, nothing above it is stable.
What blocks it: exclusion patterns, in-group/out-group dynamics, transactional relationships, remote work without deliberate connection.
A common purpose that aligns diverse experiences. People know what the team is for and why it matters. Belonging answers "am I wanted here?" — shared path answers "are we going somewhere together?"
What blocks it: competing agendas, hidden performance incentives, purpose statements that don't survive a difficult Tuesday.
Guidance, mentoring, and the permission to not know yet. In an AI-saturated workplace, learning becomes continuous. Teams without space to learn default to faking competence — which eventually breaks under pressure.
What blocks it: performance cultures that punish visible novice behaviour; leaders who model certainty instead of curiosity.
People are empowered to contribute to decisions and take ownership of outcomes. Voice without responsibility produces entitlement. Responsibility without voice produces resignation. The stage requires both.
What blocks it: consultation theatre, where input is gathered and then ignored; hierarchical bottlenecks that concentrate decision-making.
Experimentation is welcomed. People try things that might not work and report honestly on what they found. This is where safety starts crossing into growth mindset territory.
What blocks it: failure intolerance disguised as excellence; narrow KPIs that punish deviation; cultures that celebrate only success.
People feel free to question norms, including the leader's assumptions, and to challenge the status quo. This is the apex of psychological safety and the floor of growth mindset at scale.
What blocks it: status protection; leaders who confuse challenge with disloyalty; the conspiracy of silence.
Most organisations claim they operate at stage five or six. In practice, they've never secured stage one. Belonging gets assumed because nobody's visibly miserable. Shared path gets assumed because there's a mission statement. And then leaders wonder why nobody's exploring or challenging — because the foundation was never built.
In the rush to get to the exciting stages — voice, exploration, challenge — leaders routinely skip stages one and two. Belonging and shared path feel soft. They're not measurable in a quarter. They don't come up in the board report. So they get treated as prerequisites that were probably handled at some point, rather than as conditions that require constant tending.
This is a mistake. Belonging and shared path are particularly vulnerable to the ebbs and flows of relationships and situational changes. A reorganisation dents belonging. A strategy pivot undermines shared path. A leadership change resets both. When physical proximity is compromised — as it now routinely is — belonging becomes harder to sustain. When AI and constant strategic churn reshape interactions, shared path blurs.
Nurturing strong relationships and a genuine sense of collective purpose is not a stage you pass through; it's a foundation you keep reinforcing. Every single layer above it fails without it.
The six-stage model was originally developed for co-located teams. The workplace has changed. Three forces now reshape every stage:
Tom Allen's classic research established that the strength of workplace communication drops off sharply with physical distance. A colleague ten metres away gets meaningful collaboration; one sixty metres away may as well be on another continent. Global, distributed, hybrid teams routinely operate at distances the Allen Curve predicts will kill collaboration outright. Belonging has to be built deliberately — through ritual, cadence, and visible investment — rather than emerging naturally from proximity.
Volatility, uncertainty, complexity, ambiguity, and hyper-connectedness — the modern operating context. In VUCAH conditions, psychological safety is under constant pressure. Leaders can't predict outcomes, which makes empowerment feel risky. Teams can't rely on stable ground, which makes challenge feel reckless. The stages don't change, but the effort required to hold each one increases.
AI erodes psychological safety in a particular way: it introduces existential uncertainty about the value of the roles people currently hold. A person worried their job might be automated in eighteen months is not at full capacity to explore or challenge. Leaders who ignore this reality — who talk about AI as a productivity tool without acknowledging its identity implications for the team — corrode safety without realising they're doing it.
Here's what the individualistic translation of psychological safety tends to skip: the role of power.
Power and status differences are the main culprits preventing psychological safety in a team.
When people don't speak up, it is almost never because they're insufficiently brave. It's because they've correctly read the power dynamics and concluded that speaking up will cost them. The research on upward voice is unambiguous: employees calibrate truth-telling to perceived risk. In cultures where the risk is high, silence is rational. Training the silent to be brave, without changing the risk calculus, is not a solution. It's an off-loading of leadership responsibility onto the least powerful people in the room.
This is why the relational view puts leadership behaviour at the centre. Leaders set the risk calculus. They decide what happens to the first person who says something uncomfortable. They control, by example, whether challenge is met with curiosity or consequence. The leader is the single most powerful variable in whether a team is psychologically safe.
Safe2Great research over years of leadership 360 assessments shows something counterintuitive. The leaders most likely to block psychological safety are not the obvious villains — the loud, dominant, controlling-competitive ones. They're the critical-sceptical leaders: the quiet, intellectually cutting, perpetually doubting operators whose style looks like rigour but functions as corrosion.
A Hippo-style controlling leader does damage, yes — but that damage is visible, and teams develop workarounds. A critical-sceptical Clam leader does more damage because the criticism is framed as quality control, the doubt as intellectual virtue. Over time, the team learns that any new idea will be found wanting. They stop offering ideas. The leader experiences this as "my people don't bring me good thinking," unaware they've spent three years training silence.
The other revelation: the best predictors of psychological safety in teams are not the absence of controlling behaviour but the presence of relational growth mindset operating modes — specifically Lift Others Up and Team Up. Safety is built by leaders who actively do something positive, not by leaders who merely refrain from doing something negative.
The six-stage model is deliberately constructed to make this next point unmissable. Stages one through four build the floor — the conditions that allow a team to function without fear eating its capacity. Stages five and six cross into growth mindset: exploration, challenge, development.
Most organisations treat psychological safety as the destination. They build the floor, celebrate it, stop there. Then they wonder why their safe team isn't producing anything ambitious. The psychological safety plateau is real. Safety without growth becomes comfort. Comfort becomes complacency. Complacency becomes the very thing safety was supposed to unlock us from — a team too settled to challenge itself.
Growth mindset is not possible without psychological safety. Psychological safety is not enough without growth mindset.
This is the core operating principle of Safe to Great. The two concepts are not parallel virtues you balance. They're one developmental continuum you move through. Safety is necessary but not sufficient. Growth is the work that safety enables.
Psychological safety is measurable. Self-report surveys capture some of it, but they're vulnerable to the very dynamic they're trying to detect: people in unsafe teams tend to rate their teams as safe, because rating the team as unsafe would itself feel unsafe. The more sophisticated approach triangulates self-report with behavioural observation, leadership 360 data, and team-level pattern analysis.
The Safe2Great assessment suite includes four instruments that, together, map where a team is across the six stages: the Growth Mindset Leadership 360, the Growth Mindset Self Test, the Great Teams Assessment, and the Culture for Growth Survey. All benchmarked against a global database of thousands of leaders and teams. Used together, they tell you not just whether a team feels safe, but which specific stage is broken and what's blocking it.
Abstract commitment to psychological safety doesn't build it. What builds it is what leaders do in real moments of pressure. Six practical actions, one for each stage:
None of these require a new initiative, a workshop, or a consultant. They require the leader to do them repeatedly, under pressure, in front of the team, until the team believes them. Which, in most organisations, takes about two years longer than anyone expects.
"Growth mindset is not possible without psychological safety. Psychological safety is not enough without growth mindset."
Detailed explorations, theory, and practical application for practitioners who want the full picture.