In February 2025, the Pew Research Center published a survey of 5,273 employed US workers. Half of them — 52% — said they feel worried about the future impact of AI in the workplace. Another 36% said they feel hopeful. It looked like an uncomfortable tie.
But there is a third number that changes the reading: 55% rarely or never use AI tools at work. The people carrying that worry are not the ones using the technology. It is a speculative concern, not one born from experience — and that says more about how a company communicates AI than about AI itself.
Globally, KPMG and the University of Melbourne found something similar. They surveyed more than 48,000 people across 47 countries and found that 66% use AI regularly — but only 46% trust it. Adoption grew. Trust did not. There is a visible gap between doing and believing, and that gap does not close with motivational talks.
The fear has a concrete reason
Forrester asked American, British, German, French and Australian workers about their fears around AI. Forty-three percent said they believe many people will lose their jobs to automation over the next five years. Twenty-five percent suspect it will affect their own job.
But the more revealing figure is different. Fifty-one percent of UK business leaders see AI as a way to cut investment in staff. Forty-three percent expect to reduce entry-level roles. That conversation where the CEO announces "we're rolling out AI to boost productivity" gets heard by the team as "we're rolling out AI to hire fewer people." It is the same announcement, but the meaning has changed.
Deloitte put numbers to that anxiety. Among frontline workers, trust in company-provided AI tools fell 31% between May and July 2025 — and trust in autonomous AI systems, the ones that make decisions without waiting on a person, dropped 89% in that same group over the same period. When employees do not know what AI can do on its own, they assume the worst.
In Brazil, KPMG measured the same comparison on local ground. In the Brazilian cut of the study, published in 2025, 86% of respondents said the company they work for already uses AI, but only 55% said they were willing to trust decisions made by it. Nearly 80% express concern about the technology's risks, and 46% fear becoming obsolete if they do not adopt AI quickly. Adoption ran ahead of trust — and that gap, not the technology itself, is what later shows up as quiet resistance.
Why "don't use AI without permission" does not work
Some of our employees fear job loss, and it turns them away from AI altogether.
Policies written on paper do not change behavior when there is a vacuum. If someone has an urgent task today and the only tool that solves it sits outside the company's perimeter, they choose to deliver.
The fear is not irrational. But it is not about AI itself either. It is about uncertainty: who authorizes what AI does, how anyone will find out it happened, and who is accountable if something goes wrong. When those three things become clear — rule, visibility, accountability — the conversation changes.
Gartner found that most organizations are still stuck at the "governance on paper" stage. They write policies but do not implement the operational controls a team can actually see working every day. The risks of data compromise, third-party leakage and inaccurate outputs stay the same, because what changed is the document, not the workflow.
In practice, that vacuum shows up in ordinary office routines. Someone gets an urgent spreadsheet on a Friday afternoon and pastes pieces of it into a free chatbot, because that is faster than waiting for IT to approve an official tool. Nobody approved it, nobody will find out it happened, and no policy stopped it — because the policy never turned into a button people could actually use inside their own workflow.
What has to be in place
Role-based access. If a connector has a hundred functions and support needs two, release those two — not the whole tool. Access becomes function by function, and each person sees only what their job authorizes.
Approval before a sensitive action. AI stops and asks for confirmation inside the conversation. Documents that become the company's official knowledge can require two-step approval, from people with different roles — whoever writes kept separate from whoever approves.
Visible audit trail. Creating an agent, sharing a file, changing a permission, running an important action: all of it gets logged. In an audit or an incident investigation, that is the difference between reconstructing what happened in minutes and never finding out.
Corporate identity from the directory. People sign in with the same login as the rest of the network. Someone removed from the directory loses access to AI along with it — no dependency on someone remembering to revoke a stray account. Sessions expire automatically. Credentials on connected tools stay under whoever authorizes them, never shared.
Approved knowledge with sources. AI's answer is not a lucky guess. It comes from documents that have an owner, a current version and a defined scope. Critical answers can cite the source they used — because we know what that source is.
Reuse with permissions by person and role. An agent that works well, a script that got it right, a flow that saved time — all of it can be reused by other people, but within a defined scope. Someone creates it; the whole company learns from it.
That is how Skyller was designed: identity from the directory, roles that define access, approval and automatic logging as the default. The rules are visible every day, not buried in a document nobody opens.
From individual improvisation to capability that sticks
When AI use is invisible — everyone on their own personal account, with tools they paid for themselves — the knowledge that works stays locked away. The analyst who spent weeks building the perfect script for answering a tax question is the only one who has it. The person who figured out how to structure a research task with AI does not share it, because there is no official place to share it.
That fragility weighs even more in markets with high turnover, a common scenario for companies across Brazil and Latin America. When the person who mastered a process leaves the company, the knowledge leaves with them if it was never recorded somewhere the rest of the team can reach. An environment where agents, scripts and flows stay available to whoever takes over the role next turns that vulnerability into continuity: the work outlives the person who built it.
In an environment with defined access and permitted reuse, the gain changes scale. Agents, prompts, flows and spaces can be made available to other people within a defined scope. What one colleague got right becomes team capability, not one person's secret. Credits shared across the team also solve the budget problem: whoever needs more uses more, and consumption stays visible — no idle licenses on one side and people hitting limits on the other.
And the fear changes too. When people know every action leaves a trail, that approval exists for what matters, that there is a clear rule instead of guesswork, trust grows. Not trust in the technology. Trust in the company — in how it governs that tool.
Three questions to take into your next meeting
Before writing another policy about AI use, talk to IT and operations leadership. The answers will tell you whether you are on paper or in practice.
-
If an employee is let go today, how many AI tools can they still sign into tomorrow? If the answer is "it depends on someone remembering to cancel accounts," the problem is not policy — it is identity.
-
When AI answers something about an internal rule or a critical process, where did that content come from and who authorized it to be used that way? Without a review and approval step, any outdated file can become the official answer.
-
What did the company learn about its own AI use last quarter? How many people, doing what, with what results, approving what. If the answer is "nothing," usage is not lower. It is just out of sight.
