I recently facilitated an AI workshop where something revealing happened. During the hands-on exercises, several participants kept their screens angled away from colleagues. They were afraid of being seen making mistakes with the tools.

If people won’t even experiment during dedicated training sessions, what chance do we have of them experimenting in their actual work?

This experience crystallised something I’d been observing for months: psychological safety isn’t just nice to have for AI adoption. It’s the foundation everything else depends on.

What Psychological Safety Actually Means

Psychological safety is the belief that you won’t be punished or humiliated for speaking up with ideas, questions, concerns, or mistakes. Amy Edmondson’s research on this concept, widely cited in Harvard Business Review, has transformed how we think about team performance.

In the context of AI adoption, psychological safety means:

• Feeling safe to try AI tools without mastering them first
• Asking “basic” questions without embarrassment
• Sharing failed experiments as learning opportunities
• Admitting when you don’t understand something
• Voicing concerns about AI without being labelled resistant

Without this safety, people either avoid AI entirely or use it secretly, neither of which serves organisational goals.

Why AI Uniquely Threatens Psychological Safety

AI creates psychological safety challenges that other technology adoptions don’t:

Competence Identity Threat

Many professionals have built their identity around being competent. AI tools can produce outputs that took years to learn to create. This threatens the competence that defines professional identity.

When competence feels threatened, people protect themselves by avoiding the threat—which means avoiding AI.

Visible Capability Gaps

Using AI tools often reveals capability gaps publicly. Struggling with prompts, generating poor outputs, not knowing what to ask—these happen in real-time, potentially visible to colleagues.

The risk of public incompetence keeps people from trying.

Generational Dynamics

AI often inverts typical expertise hierarchies. Younger employees may adapt faster than experienced professionals. This dynamic creates anxiety for senior people and awkwardness for junior ones.

Neither group benefits from this tension.

Job Security Anxiety

AI conversations inevitably trigger job security concerns. Even if an organisation has no plans for workforce reduction, people worry. That worry makes genuine engagement with AI feel risky.

Signs of Insufficient Psychological Safety

How do you know if psychological safety is lacking for AI adoption? Watch for:

Silence in training sessions. People don’t ask questions or admit confusion. Sessions feel more like compliance exercises than learning experiences.

Private experimentation only. Some people use AI tools, but only when no one’s watching. They don’t share what they’re learning.

Blame-shifting when things go wrong. When AI outputs cause problems, people focus on assigning blame rather than learning from the experience.

Resistance framed as practical concerns. People express resistance through seemingly practical objections (“it’s not ready,” “it’s not relevant”) rather than admitting underlying fears.

Champions working in isolation. AI enthusiasts don’t share their work because they fear being seen as showing off or making others look bad.

Building Psychological Safety for AI

Creating psychological safety requires deliberate action at multiple levels.

Leadership Modeling

Leaders set the tone. When leaders:

• Admit their own AI learning struggles publicly
• Share their failed experiments along with successes
• Ask “basic” questions without embarrassment
• Respond to mistakes with curiosity rather than blame

They signal that these behaviours are safe for everyone.

I worked with a director who started every team meeting by sharing “my AI failure of the week.” It completely changed how her team engaged with AI experimentation.

Framing AI Adoption as Learning

The frame we put around AI adoption matters enormously.

Threatening frame: “You need to learn these tools to remain valuable.”

Safe frame: “We’re all learning together. The goal is exploration and discovery, not immediate mastery.”

Same content, very different psychological impact.

Position AI adoption as a collective learning journey. Normalise the learning curve. Celebrate exploration, not just successful outcomes.

Creating Safe Practice Spaces

People need spaces to practice without performance pressure:

• Dedicated learning time separate from production work
• Sandbox environments where mistakes don’t matter
• Peer learning groups where everyone is at similar levels
• Anonymous question channels for sensitive queries

The more separate practice feels from performance evaluation, the more people will actually practice.

Responding Constructively to Failures

How the organisation responds when AI experiments go wrong determines whether people keep experimenting.

Destructive responses:

• “Who approved using AI for this?”
• “This is why we should be careful with these tools.”
• “We need more controls on who can use AI.”

Constructive responses:

• “What did we learn from this?”
• “How can we experiment more safely next time?”
• “This is exactly the kind of learning we need.”

The first set of responses guarantees people stop experimenting. The second set encourages continued learning.

Addressing Job Security Directly

You can’t create psychological safety about AI while leaving job security concerns unaddressed.

Be direct:

• If AI will affect roles, say so with specifics
• If it won’t, make that commitment clearly
• If you’re uncertain, acknowledge the uncertainty

Vagueness allows worst-case-scenario thinking to flourish. Honest communication—even when the news is uncertain—builds more trust than corporate platitudes.

Team-Level Practices

Psychological safety is built largely at the team level. Practices that help:

AI Learning Circles

Small groups that meet regularly to share AI experiences—successes, failures, questions. The regularity normalises ongoing learning. The small size creates intimacy.

Failure Retrospectives

When AI experiments don’t work, hold retrospectives focused on learning rather than blame. What happened? What did we learn? What will we do differently?

Pairing and Shadowing

Pair people to experiment together. Shared learning feels safer than individual exposure. Plus, two people often figure things out faster than one.

Question Walls

Create physical or virtual spaces where anyone can post AI questions anonymously. Others can add answers or admit they have the same question. Anonymity removes fear of looking ignorant.

Celebration of Learning

Recognise and celebrate learning effort, not just successful outcomes. “Sarah spent three hours this week experimenting with AI for report generation” deserves recognition even if the results weren’t production-ready.

What Managers Must Do Differently

Managers play a decisive role in psychological safety. Specific behaviours that matter:

Ask questions first. Before offering answers or solutions, ask questions. This positions learning as the goal, not performing.

Share your own struggles. Managers who appear to have AI figured out create pressure. Managers who share their own learning journey create permission.

Respond to mistakes with curiosity. The first words after something goes wrong set the tone. “What happened? What can we learn?” rather than “How did this happen? Who’s responsible?”

Protect experimentation time. If learning time keeps getting cancelled for “real work,” people learn that experimentation isn’t actually valued.

Be specific about what safe means. Abstract safety commitments don’t land. “If an AI experiment goes wrong, we’ll treat it as learning, not performance” is more powerful than “we support experimentation.”

The Paradox of High-Performing Teams

Here’s something counterintuitive: high-performing teams often have less psychological safety for AI than lower-performing teams.

Why? High performers have succeeded by being good at current approaches. AI threatens to devalue that competence. The higher the investment in current skills, the greater the threat from new ones.

This means your best people may need the most psychological safety support around AI adoption. Don’t assume high performers will just figure it out.

Measuring Psychological Safety

How do you know if your efforts are working? Some indicators:

• Number of questions asked in training sessions
• Willingness to share experiments (successes and failures)
• Requests for learning resources and support
• Speed of adoption across different teams
• Quality of discussion in retrospectives

Surveys can help, but observed behaviour often tells more than self-reported attitudes.

The Long Game

Psychological safety isn’t built in a workshop. It’s built through consistent behaviour over time.

Every interaction either adds to or subtracts from the safety bank. A single blame-heavy response to a failure can undo months of safety-building effort.

This means psychological safety requires ongoing attention, not a one-time initiative.

The Foundation for Everything Else

All the AI training, tools, and resources in the world won’t drive adoption if people don’t feel safe using them.

Psychological safety is the foundation. Build it first.

Then build everything else on top.