When someone goes quiet in a meeting about AI, asks the same question about job security for the third time, or suddenly becomes the loudest champion in the room, they're not malfunctioning. They're communicating. Managers who read behaviour well catch a stall long before it shows up in the numbers, because behaviour is data.
Most managers already notice these moments. What's harder is resisting the pull to respond to the surface behaviour rather than the signal underneath it — and a missed signal doesn't disappear, it resurfaces later, as a workaround or a quiet drop in use. That happens by:
Research from Writer and Workplace Intelligence in 2026 found that 29% of employees admit to actively undermining their organisation's AI rollout, and 44% of Gen Z workers say the same — close to a third of them citing fear of losing their job as the reason. That's not defiance. It's the same behaviour this section teaches you to read, just further along the same line.
Here's what to look for — and what each signal is actually telling you:
Next time you catch one of these signals, pause before responding. The question underneath it is usually more useful than the one on the surface.
Resistance to AI tends to cluster around five recognisable signals, not an endless range of individual quirks. Each one is legitimate, and each one points to something you can actually do something about.
They map to SCARF, a model developed by David Rock at the NeuroLeadership Institute in 2008, which names five domains people stay alert to whenever change is underway:
SCARF tells you which domain is under threat. What you build in response is sometimes described as a SAFETY approach — creating security, autonomy, fairness, esteem, trust and room for someone's own context. Read the "what they need" box for each domain to understand how to use the SAFETY principles.
Open each domain below to see what it looks and sounds like during an AI rollout, and what the person sending the signal actually needs from you.
"Am I still the expert here?"
This shows up as the experienced team member who becomes critical of AI outputs, or the senior analyst who suddenly becomes very interested in "quality standards." They're not being obstructive — they're protecting something real, because their expertise is their identity. When AI can do what they do, they need to know they still bring something irreplaceable.
"What does this actually mean for my role?"
This shows up as repeated questions about what will change, a check-in with HR, or asking what "using AI" will actually mean for their performance review. Not knowing what the future looks like is deeply uncomfortable, and the gap tends to get filled with worst-case scenarios.
"Was I consulted?"
This shows up as people finding workarounds, using the tool in their own way, or quietly continuing to do things as they always did without openly resisting. Autonomy concerns aren't about being difficult — they're about agency: the tool arrived, nobody asked, and now they're expected to use it.
"Is this being handled consistently?"
This shows up as questions about whether senior staff are also expected to use AI, noticing when some people get more support than others, or frustration when the rules feel unclear or inconsistently applied. Perceived unfairness spreads fast and poisons momentum.
There's a version of the AI conversation that creates safety, and a version that deepens the threat response. The difference is rarely what you say — it's how you start.
This isn't a one-off trick for AI conversations specifically — it's the same principle behind change work generally. Resistance tells you something your plan missed, whether that's a fear, a point of confusion, or a genuine problem with the rollout itself. Treat it as information, and a difficult conversation turns into something useful — for the plan as much as for the person.
Conversations that shut things down position you as someone delivering news, not someone interested in their experience — and they tend to open the same way:
The goal of the first conversation isn't to solve anything. It's to make it safe to be honest. If someone leaves the conversation feeling heard, you've done your job.
The conversations that open things up look different: start with curiosity ("I've been thinking about what this means for your work — I'd love to understand what it's been like from your end"), name what you've noticed without labelling it ("I've noticed you've been quieter in the AI sessions — I wasn't sure if that was landing well"), make space for the honest answer ("You don't have to be positive about this. I'm more interested in what's actually true for you right now").
By this point you've had the conversation more than once. You've listened, acknowledged, given clarity, choice and space — and someone is still not moving. This section is about what to do next, without forcing the issue or quietly giving up on it.
First: distinguish between resistance and refusal. Resistance is a signal that something hasn't been addressed yet. Refusal is a choice. Most of what looks like refusal is actually still resistance — the person hasn't yet felt heard enough, or safe enough, or clear enough. Before treating something as refusal, check whether you've genuinely named and responded to the specific concern driving it.
How to diagnose which one you're dealing with:
Resist the urge to force it. The same 2026 research cited in Section 1 found that a majority of C-suite leaders are already planning layoffs for employees who don't adopt AI fast enough — exactly the kind of pressure that tips resistance into sabotage instead of resolving it.
When it is genuine refusal, be clear about expectations without ultimatums, then hold the line with warmth. The combination of clarity and care is what makes that conversation land differently than a performance conversation.
AI rollouts create natural divisions: early adopters who run ahead, resistors who dig in, and a large middle group who are watching both to decide what to do. Managing all three at once is one of the hardest parts of this change.
This has been confirmed in research across the wider workforce, too. Indeed's 2026 AU Jobs & Hiring Trends Report found the same three-way split: 32% of Australian workers already use AI multiple times a week, another 32% have disengaged from it entirely, and the remaining third sit uncommitted in the middle. The disengaged third didn't reject AI — they were never given the structure to use it well, which is exactly why the middle group here is worth the effort.
The temptation is to focus on the resistors — they're the loudest signal. But neglecting the early adopters creates resentment ("I've been doing this for months, why isn't everyone else expected to?"), and ignoring the middle group means losing the majority of your team to inertia. The quiet majority is where most adoption is actually won or lost.
The same 2026 Writer and Workplace Intelligence research cited in Section 1 also tracked what happens once someone becomes a genuine super-user: they save roughly nine hours a week against two hours for slower adopters, and they're around three times more likely to have received a promotion or pay rise. That gap compounds every week it's left alone — the clearest argument for closing it early rather than letting the middle group drift.
"Use your early adopters as translators, not evangelists." The difference matters:
One is testimony and the other is a use case someone can actually copy — that's the real difference between inspiring people and moving them.
How to move together as a team:
The goal is a team culture where using AI is normal — where it's part of how the team works, full stop.
Reading Your Team
Your five answers, gathered from each section above. Edit them up there — this is a read-only copy for review, saving and printing.