For two years, the working assumption among HR leaders rolling out generative AI has been that fear fades with familiarity: give people hands-on time with the tools, and the anxiety about being replaced will settle down as the technology starts to feel normal. New research says the opposite is happening, and I think that should change how leaders are approaching AI rollout right now.
idealis, a workplace research and advisory firm, published its follow-up “Ready or Not, It’s Here: AI Adoption at Work 2026” study in August, surveying more than 15,000 US workers through CivicScience. The headline number should stop any HR leader mid-scroll: 72 percent of employees who actively use AI in their jobs worry about their own job security, compared with 46 percent of employees who do not use it. Sixty-two percent of workers now use AI professionally, up 16 percentage points from 2025. Adoption is climbing fast. Comfort is not following it.
The counter-argument, and why it does not hold
The obvious objection is that this is just a lag effect: adoption always outpaces comfort in the early stages, and the discomfort will catch up and fade as AI use becomes routine, the way it did with spreadsheets or email. I understand the instinct. It is the same logic that has justified a lot of “just get people using it” rollout strategies over the past two years.
But the idealis data does not describe a lag. It describes an inversion. If familiarity were doing what leaders assume it does, the heaviest users of AI should be the least anxious group in the survey, not the most. Instead they are the most anxious. Leaders themselves, who by the study’s own numbers use AI at a higher rate than individual contributors (76 percent versus 61 percent), are also more worried about AI’s proliferation than the people they manage, 88 percent versus 76 percent. The people closest to the technology, at every level of the org chart, are the most unsettled by it. That is not a comfort gap that closes on its own with time. It is a trust gap that a rollout plan has to actively close.
What the fear is actually about
I do not think this data means employees are irrational about AI, or that the tools should be pulled back. I think it means employees using AI daily can now see exactly which parts of their own job the tool has gotten good at, and nobody has told them what happens next. Vague reassurance that “AI will augment, not replace” does not survive contact with a worker who just watched a model do a task that used to take them an afternoon in ninety seconds.
Dr. Sumona De Graaf, founder and CEO of idealis, put the responsibility squarely on leadership rather than on the workforce’s adaptability: “The hard part is setting the guardrails, providing training, and earning trust from the top.” That framing matters. It is not a call for more AI literacy training alone. It is a statement that the guardrails, and the trust, have to come from leadership decisions, not from employees eventually getting used to the tool.
What HR leaders should actually do
Three things follow from this, and none of them are “slow down AI adoption.”
First, pair every AI rollout with an explicit statement of what happens to the people whose tasks it touches: retraining paths, redeployment commitments, or, where reductions are genuinely necessary, a clear and honest timeline instead of silence. Trust gaps inside hiring teams have already shown what happens when leaders deploy AI tools faster than they explain them: adoption stalls even when the tool works.
Second, stop treating heavy AI users as the group that needs the least attention in a change management plan. The data says they need the most. They are the ones who can see the tool’s capability most clearly, and their anxiety is a rational read of what they are watching, not a training gap to be closed with a tutorial.
Third, be skeptical of any vendor pitch, or any internal business case, that assumes productivity claims will translate cleanly into morale, retention, or trust. As I have argued before, vendor AI productivity numbers deserve scrutiny on their own terms, and this data adds a second reason: even a genuinely productive rollout can still erode the workforce’s confidence if leadership never addresses what the productivity gain means for the people who produced it.
The idealis numbers are not an argument against AI adoption. They are evidence that the leadership work of adoption, the guardrails, the training, the honesty about consequences, has been running well behind the technical rollout. Until that gap closes, giving employees more hands-on AI time will keep producing more anxious employees, not fewer.
Source: idealis
