Why the freedom to say “I don’t understand this yet” protects your people, your adoption and your results.
AI is changing how work gets done. Beneath the visible shift in tools, processes and skills, another transition is taking place. People are working out what they are expected to know, what they can safely question, and what happens if they admit they are not yet sure.
For high-functioning leaders, that uncertainty can be particularly difficult to voice. You are used to being the person who knows. The person others turn to. Saying “I don’t understand this yet” can feel less like a learning moment and more like an admission that you are falling behind. So the question stays unasked.
Psychological safety gives people enough confidence to bring that uncertainty into the room, where a team can work with it rather than leaving individuals to carry it alone.
Key takeaways
- Psychological safety is the shared belief that a team is safe for interpersonal risk, such as asking a question, admitting a mistake or raising a concern.
- Research links psychological safety with learning, performance and lower levels of emotional exhaustion, although the evidence does not show that psychological safety alone causes better outcomes.
- During the AI transition, uncertainty can become a hidden source of cognitive and emotional load when people feel unable to question, challenge or admit what they do not yet understand.
- Recent research suggests that the way organisations introduce and manage AI matters to people’s experience of work, including psychological safety and cognitive load.
- Leaders can strengthen psychological safety through small, repeated behaviours: going first, welcoming early questions, separating problems from blame, clarifying what AI will and will not change, and designing AI work within human cognitive limits.
Consider a seasoned chief financial officer who privately re-checks every AI-generated forecast by hand at 11pm. She does not trust the new model, yet feels unable to say so without appearing resistant to change or behind the curve. Three weeks pass before anyone notices her exhaustion. The cost of an unasked question rarely appears on a balance sheet, but capable people tend to carry it longest.
During the AI transition, the question worth asking is what makes that CFO feel she has to work this way in the first place. If people can acknowledge uncertainty early, leaders have an opportunity to address the underlying issue before it becomes additional work, private anxiety or sustained strain.
What is psychological safety, and why does it matter?
Psychological safety describes a shared belief that a team is safe for interpersonal risk. In practical terms, people can ask, admit, challenge and disagree without believing that doing so will damage their standing.
The concept was developed by Harvard’s Amy Edmondson, whose 1999 study of 51 work teams in a manufacturing company found that psychological safety was associated with learning behaviour, which in turn related to team performance. Her work helped establish an important distinction: a psychologically safe team can maintain high standards while making room for questions, mistakes and difficult information.
The wider evidence points in a similar direction. A meta-analysis by Frazier and colleagues, covering 136 independent samples, more than 22,000 individuals and nearly 5,000 groups, found psychological safety related to outcomes including task performance and organisational citizenship behaviour. Google’s Project Aristotle later identified psychological safety as the most important of five dynamics Google associated with effective teams.
The value of psychological safety becomes particularly clear when the work itself is changing. New technology creates unfamiliar decisions, new expectations and gaps in knowledge. People need somewhere to put those uncertainties before they can learn their way through them.
What does psychological safety have to do with burnout?
The World Health Organization defines burnout as an occupational phenomenon resulting from chronic workplace stress that has not been successfully managed. That definition keeps the focus on the conditions in which work takes place, alongside the individual experience of strain.
Psychological safety forms part of that wider environment. When people can raise concerns, question decisions and ask for clarification, teams have more opportunity to address sources of pressure while they are still manageable. When those conversations feel risky, individuals may compensate privately instead.
That can create a hidden second job: doing the work while also managing the impression that everything is under control. Someone checks an AI output repeatedly because they do not feel able to question the system. Another person spends the evening teaching themselves a tool they were afraid to admit they did not understand. A third stays silent in a meeting, then spends hours working through the issue alone.
Over time, those small acts of compensation can consume capacity. The shift from running on reserves to leading from vitality rarely happens because of one difficult moment. More often, strain accumulates through the ordinary demands of work and the effort required to keep absorbing them.
Why does AI put additional pressure on psychological safety?
AI tools arrive quickly and change frequently. For experienced professionals, that can create an uncomfortable gap between their established expertise and their emerging competence.
A senior leader may be highly confident in finance, operations, strategy or people leadership while still learning how to evaluate an AI-generated recommendation. Admitting that gap can feel surprisingly exposing, particularly in cultures where competence has become closely associated with having an answer.
The result may look like resistance when the underlying issue is uncertainty.
Research is beginning to illuminate this dynamic. A 2025 three-wave study of 381 employees in South Korean companies by Kim, Kim and Lee found that organisational AI adoption was associated with lower psychological safety. Lower psychological safety was also associated with higher depression, and psychological safety statistically mediated the relationship between AI adoption and depression. The study found that ethical leadership moderated the relationship between AI adoption and psychological safety.
The findings do not establish that AI adoption causes depression. They do, however, suggest that the human experience of adopting AI deserves attention alongside the technology itself.
A separate 2025 survey by MIT Technology Review Insights of more than 500 executives and business leaders found that 22% reported having hesitated to lead or suggest an AI initiative because of fear of failure or criticism. Sixty percent identified clarity about how AI would and would not affect jobs as something that could improve psychological safety, while 51% pointed to leadership openness to questions, dissent and failure.
These findings come from a survey of executive perceptions rather than an experiment, so they cannot establish cause and effect. They do offer a useful glimpse into the leadership climate around AI adoption: people are making judgements about what feels safe to say, and those judgements influence how openly they engage with change.
What happens when uncertainty stays unspoken?
When uncertainty has nowhere to go, people tend to solve around it.
They check more than necessary, compensate for unclear processes, avoid asking questions that might reveal a knowledge gap, and take problems home because it feels easier than exposing them in the room.
Research in other high-pressure environments suggests that psychological safety can influence how those demands are experienced. A 2024 cross-sectional study of 621 nurse practitioners in California found that higher psychological safety was associated with lower emotional exhaustion and depersonalisation. It also found that psychological safety weakened the relationship between aspects of the work environment and burnout indicators.
The study does not prove that psychological safety prevents burnout. Its value lies in showing how the social conditions around demanding work can shape the relationship between the work environment and people’s experience of exhaustion.
For leaders, that distinction matters. It shifts the conversation away from asking how much more pressure a capable person can absorb and towards understanding what the work is asking them to carry.
AI creates a cognitive load of its own
The AI transition also introduces a different kind of strain: the mental effort involved in monitoring, checking and making decisions around AI-generated work.
A 2026 Harvard Business Review study of 1,488 full-time US workers, conducted by Boston Consulting Group researchers, described “AI brain fry” as mental fatigue associated with excessive use or oversight of AI tools beyond people’s cognitive capacity. The researchers found associations with more errors, decision fatigue and intention to quit.
They also made an important distinction between this mental fatigue and burnout. When workers used AI to replace routine or repetitive tasks, they reported lower burnout scores, although that did not necessarily reduce mental fatigue.
That finding is worth sitting with.
The same technology can reduce one form of work while introducing another. An AI tool might remove hours of repetitive analysis yet require constant checking of its output. It might accelerate the production of information while making it harder to determine which information deserves attention.
The leadership question, then, becomes more interesting than whether people are using AI. It is whether the new way of working is giving them greater capacity or simply moving the effort somewhere less visible.
For a deeper look at how cognitive overload affects judgement, see decision fatigue and leaders’ judgement quality.
Why is psychological safety a business priority during AI adoption?
For HR and people leaders, psychological safety belongs in the conversation about performance, risk and adoption because people cannot respond intelligently to change if important information remains unspoken.
A leader with enough capacity can pause, assess what is happening and make a considered judgement. They are also better placed to notice when the team is struggling with something that has not yet reached the formal agenda.
That steadiness has a wider effect. Think of a conductor. They do not play every instrument, but their tempo gives the orchestra something to work from. In an organisation, the way a leader responds to uncertainty helps establish what other people believe they can safely say.
The MIT Technology Review Insights survey found that 83% of executives said psychological safety has a measurable impact on the success of AI initiatives. Again, this represents what those executives reported rather than experimental evidence that psychological safety causes AI success. It is nevertheless a significant signal from leaders who are already navigating the transition.
Clarity also matters. The same survey found that executives saw greater clarity about how AI would and would not affect jobs as an important way to strengthen psychological safety.
People can work with uncertainty when they know where they stand.
That is why burnout prevention and successful AI adoption can sit within the same leadership conversation. Both depend, in part, on whether people have enough clarity, capacity and permission to surface what is happening before it becomes harder to address.
The aim is simple: high-functioning, fully resourced.
How can leaders build psychological safety during the AI transition?
Psychological safety grows through ordinary moments of leadership. Five behaviours can make a meaningful difference.
1. Go first
Name your own learning gaps before anyone else has to.
A senior leader who says, “I don’t yet understand how this system reached that answer,” demonstrates curiosity without pretending to have all the answers. That small admission changes the social cost of learning in public.
2. Reward the first question
The first response to a difficult question tells everyone else something about what happens when they speak up.
Meet uncertainty with curiosity. Thank people for surfacing concerns, even when you disagree with the conclusion. A question does not have to be correct to be useful.
3. Separate the problem from the person
When an AI output is wrong, begin by understanding what happened.
What did the team expect? Which assumption failed? What information was missing? What needs to change?
This approach keeps attention on the work rather than making the person who raised the issue responsible for the discomfort it creates. Problems can then become information that improves the system.
4. Be explicit about what AI will and will not change
Uncertainty becomes harder to manage when it extends beyond the technology into questions about role, status and job security.
Explain what AI is being introduced to do. Clarify where human judgement remains essential. Be honest about what has not yet been decided.
Clarity will not remove every concern, but it gives people something solid to work with.
5. Design AI work within cognitive limits
More information does not automatically create better decisions.
Use AI to reduce repetitive work where it genuinely reduces load. Define how and when outputs will be reviewed. Protect periods of focused, tool-free thinking. Pay attention to how much monitoring and decision-making the new workflow adds.
The goal is useful AI use, with enough human capacity left for judgement.
To explore whether uncertainty is going unspoken in your team, read how to tell whether your team is hiding AI uncertainty.
For scripts and routines for each of these five moves, read five leader moves, in practice.
From running on reserves to leading from vitality.
Take the Stress Continuum Assessment to see where you sit on the Stress-to-Burnout Continuum, or get in touch.
