Psychological Safety in AI Projects: 3 Practices for Project Leaders

AI is moving faster than human speed.

In Australia, most mid-sized organisations have already adopted some form of AI (either officially or under the radar). At the same time, a growing number of people report feeling overwhelmed by the pace and volume of work, with burnout becoming increasingly common.

So, if your leadership team is pushing hard on AI adoption as part of a broader AI transformation, but your project team is getting quieter, more reactive, or less willing to speak up, that is not resistance.

It is a psychological safety in the workplace problem.

Here are three grounded practices project leaders can experiment with to protect their teams, while staying realistic about what people can actually handle during rapid AI change.

1. Name the Gap Between Promise and Reality

It always helps when you name the problem.

There is often a widening gap between what AI promises and what project teams are actually experiencing on the ground. Many of the most common AI adoption challenges do not come from the technology itself, but from the additional, unacknowledged work it creates.

Workplace psychologists have already raised concerns that rapid AI rollout, without safeguards, risks eroding trust, increasing safety issues, and worsening mental health. This is not fear-mongering. It is an observation.

Here is what that gap looks like in practice:

A sponsor says, “AI will accelerate our delivery.”

At the same time, the project team is spending evenings learning prompts, reworking outputs, and doing additional quality assurance work that was never scoped.

No one says anything because they do not want to look slow, negative, or anti-innovation.

We see this ourselves. AI can be genuinely helpful for document creation, giving a faster starting framework than a blank page. But it can also create different work. Formatting is wrong. Assumptions are off. Sometimes information is simply made up.

One experiment worth trying is to explicitly surface this in your regular checkpoints:

“AI – what is helping, and what is adding noise?”

A small but practical prompt habit helps here too:

“Do not make things up. If information is missing, insert a placeholder and note what is required.”

Do not let the gap sit in silence. Silence is where frustration and quiet disengagement grow.

2. Make It Safe to Say, “I’m Not Sure This Is Right”

AI adoption is behaviour change, not just a software implementation. This is where AI change management often succeeds or fails.

When people feel they must look competent at all times, they stop asking questions. That is when errors slip through and disengagement sets in.

A small but telling example: I sat next to someone on a plane who fed a letter from Telstra into an AI assistant and asked it to explain what it meant – without reading it themselves. Telstra might need to write clearer letters, but this is where we are. People are increasingly trusting AI to interpret things they do not understand.

In projects, this shows up when someone notices an AI-generated summary is subtly wrong – the tone is off, a dependency is missing – but they do not challenge it because AI is being treated as the new standard in AI in decision making.

One simple intervention is to ask, in checkpoint meetings:

“What feels unclear or risky here that we are not naming?”

Then protect the first person who speaks. No blame. No defensiveness. Psychological safety workplace norms are built one honest contribution at a time.

3. Pair Every AI Activity with a Stop-Doing Decision

If you introduce AI but remove nothing, you have not improved productivity.

You have just increased cognitive load.

AI gets added, but meetings stay the same. Reporting stays the same. Documentation expectations stay the same. And now the team also has to learn how to prompt well, review outputs critically, and work out what to trust.

The finish line keeps moving.

We experienced this when using AI to generate first drafts of test scripts. It was only genuinely helpful after investing time understanding what inputs produced useful outputs. Otherwise, the time spent rewriting negated any gain.

That is not time saved. That is time shifted.

A practical rule that helps:

Every new AI activity must come with a stop-doing decision.

One report.

One meeting.

One manual step that is explicitly removed.

Where possible, have leadership approve that trade-off. That is how you protect capacity without pretending humans have infinite bandwidth.

A Final Thought

If AI adoption is happening at speed – and it is – psychological safety is not a soft extra. It is a delivery advantage.

Your role as a project leader is to:

Name the gap between promise and reality

Make it safe to question AI outputs

Protect capacity through real trade-offs

That is how teams keep thinking, not just coping.

If this resonates, and you are navigating the tension between AI expectations and protecting your people, that is a conversation worth having.

Retail improvement, made practical.
Leadership thinking that drives change.

Sign up to receive new articles and strategic guidance.