By Dr. Tess Breen, Organizational Psychologist

Key takeaway: AI tools are now embedded in most knowledge work, but new research from Harvard Business School professor Amy Edmondson and others shows the rollout is quietly eroding psychological safety on teams. Employees who feel uncertain about how AI judges their output go quiet, hide mistakes, and stop asking the questions that drive learning. The fix is not more training. It is a deliberate change in how leaders frame AI conversations.

Something strange is happening in the teams I work with. The smartest people, the ones who built reputations on candid feedback and curious questions, have started to go quiet. They use AI tools every day, often more than their colleagues realise, but they have stopped talking openly about how those tools are reshaping their work. When I dig in, the same pattern keeps surfacing: people are afraid of looking incompetent for needing AI, afraid of looking lazy for using it, and afraid of admitting they no longer fully understand how their own outputs are produced.

This is a psychological safety problem, and it is one of the most under-discussed leadership issues of 2026. A February 2026 piece in Harvard Business Review (Edmondson and Mortensen, “How to Foster Psychological Safety When AI Erodes Trust on Your Team”) laid out the mechanism clearly. When AI systems are opaque, when employees cannot tell whether their manager values their thinking or the model’s, and when status hierarchies start to depend on who is “good with the tools,” the foundation of speaking up cracks. People retreat into self-protection.

What the research actually says

Edmondson’s original 1999 work in Administrative Science Quarterly defined psychological safety as a shared belief that the team is safe for interpersonal risk-taking. That definition was tested on teams using telephones and whiteboards. The 2025 update, refined with Connie Hadley and Mark Mortensen for the Fearless Organization benchmark, adds new items specifically built for hybrid and AI-augmented work. The pattern in the early data is striking. Teams report similar overall safety scores as a few years ago, but the variance has widened. Inside the same company, one team is thriving and the next is deteriorating, and the differentiator is almost always how the manager talks about AI.

A 2025 study summarised by the NeuroLeadership Institute went further. Researchers found that ambiguous AI use policies (the “figure it out yourselves” approach many companies have defaulted to) drive a measurable drop in voice behaviour, the willingness to raise concerns or suggest improvements. Voice behaviour is one of the strongest leading indicators of innovation. When it falls, the consequences show up six to nine months later in product quality, customer complaints, and retention numbers.

Why the standard playbook isn’t working

Most leaders I coach have read the safety literature and believe they’re doing the right things. They run anonymous engagement surveys, they say “there are no stupid questions,” and they tell their teams it’s okay to experiment. None of that is wrong. But none of it addresses the specific kind of fear AI is generating, which is the fear of being seen as redundant. You can’t survey-anonymity your way out of someone wondering whether their job will exist in 18 months.

The shift required is from reassurance to transparency. Reassurance (“don’t worry, your job is safe”) is cheap and almost always backfires because no leader can credibly promise it. Transparency (“here is what we are testing, here is what we don’t yet know, here is how I’m using AI in my own work”) is harder but actually moves the needle. When the most senior person in the room admits they’re using a tool to draft a first version of a paper, the permission cascades downward almost instantly.

Three things leaders can do this week

The first is to run a five-minute “AI honesty round” in your next team meeting. Ask each person to share one task where they’re now using AI and one where they’re refusing to. Don’t editorialise the answers. The simple act of normalising the conversation collapses the secrecy that has built up.

The second is to separate process feedback from output feedback. When you praise a polished document, name what the person added on top of the model. “The framing in section two is yours, and it’s the strongest part” tells the team that human judgement is still what gets noticed. Without that signal, people start to assume their value is being eroded with every tool upgrade.

The third is to publicly share one mistake or limitation of your own AI use in the past week. Edmondson’s research is unambiguous on this point: the leader who admits fallibility first sets the ceiling on team candour. If you never admit anything, your team won’t either, no matter what your policies say.

The competitive advantage hiding in plain sight

In my consulting practice, the leaders who are pulling ahead right now are not the ones with the best AI stack. They’re the ones whose teams will tell them what’s broken about it. That information asymmetry, the willingness of employees to flag failures, edge cases, and ethical concerns, is becoming the rarest commodity in knowledge work. Teams with high psychological safety have it. Teams without it are flying blind into the most consequential technology transition of our careers.

If you’d like a structured conversation about how psychological safety is showing up (or disappearing) on your team, you can work with me here. I run diagnostics and coaching engagements for executives and leadership teams who want to move beyond surface-level culture work.

Frequently Asked Questions

What is psychological safety in simple terms?

Psychological safety is the shared belief on a team that you can speak up, ask questions, admit mistakes, or disagree without being punished, embarrassed, or judged. Amy Edmondson coined the term in 1999, and decades of research show it’s one of the strongest predictors of team performance, learning, and innovation.

How does AI specifically erode psychological safety?

AI introduces three new fears at work: fear of looking incompetent for needing AI, fear of looking lazy for using it, and fear of becoming redundant. When teams don’t talk about these fears openly, people withdraw, hide their AI use, and stop raising concerns. The result is a quiet drop in voice behaviour that shows up months later in quality and retention metrics.

What is the fastest way to rebuild psychological safety on a team using AI?

Start with leader transparency. The most senior person in the room should share specifically how they’re using AI, what they’re refusing to use it for, and one limitation or mistake from the past week. Research consistently shows team candour is capped at the level of the leader’s own openness, so this single behaviour has outsized impact.

Is psychological safety the same as being “nice” or avoiding conflict?

No. Edmondson is emphatic on this. Psychologically safe teams have more disagreement, not less, because people feel safe enough to challenge ideas openly. The opposite of psychological safety is silence, not conflict. A team where everyone agrees is usually a team that’s quietly afraid.

How can leaders measure psychological safety on their team?

The most rigorous tool is Edmondson’s seven-item Team Psychological Safety Scale, which takes about five minutes to complete and has been validated across industries. For a faster pulse, ask three questions in your next one-to-one: when did you last disagree with me, what’s something you’ve been hesitant to bring up, and what’s a question about AI you’ve been afraid to ask? The pattern in the answers tells you everything.

Sources

• Edmondson, A. C., & Mortensen, M. (2026). “How to Foster Psychological Safety When AI Erodes Trust on Your Team.” Harvard Business Review, February 2026. Link
• Edmondson, A. C. (1999). “Psychological Safety and Learning Behavior in Work Teams.” Administrative Science Quarterly, 44(2), 350-383.
• NeuroLeadership Institute (2025). “Latest From the Lab: Creating Psychological Safety for Improved Performance.” Link
• UNLEASH (2025). “Amy Edmondson: High psychological safety means a high learning quotient.” Link
• Fearless Organization (2025-2026). Psychological Safety Benchmark, refined with Edmondson, Hadley, and Mortensen.