Securafy AI Lab

Employee Trust in AI at Work: Why Skepticism Is Rising

Written by Jillian O. | Nov 16, 2026, 1:00:00 PM

If your team has been using AI tools for a year or more, why do they seem less convinced by them now than when they started? It is a fair question, and the 2026 research says you are not imagining it. Gallup's July 2026 Bentley-Gallup findings show the share of U.S. adults who consider themselves knowledgeable about AI rising from 64% in 2024 to 70% in 2026, while the share who say AI does more harm than good jumped from 31% to 39% in a single year. Familiarity is rising. Employee trust in AI at work is not rising with it.

The direct answer: more AI use is making employees more skeptical because everyday use exposes them to AI's errors without giving them a reliable way to judge when to trust it. People learn that the tool can be wrong, but not how to tell when. The result is a general wariness that does not translate into careful use.

The instinct for most business owners is to treat that skepticism as the problem to be solved. It is not. Skepticism is the raw material of good AI use. The real problem is that your team's skepticism is pointed in the wrong direction.

What Does the 2026 Research Actually Show About Employee Trust in AI?

It shows use and knowledge climbing while confidence falls, across several independent studies. The SCSP-Gallup American Priorities for AI study, based on a June 2026 survey of 3,152 U.S. adults, found self-reported AI knowledge rising from 60% to 68% in a year, while trust in AI remained low and largely unchanged and optimism about AI's effect on productivity and jobs declined.

Trust in how organizations handle AI is moving the wrong way too. In the same Bentley-Gallup data, the share of Americans who trust businesses "a lot" or "some" to use AI responsibly fell from 31% in 2025 to 27% in 2026, and among adults aged 18 to 29 it dropped from 30% to 20%. Bentley University's summary of the Business in Society report puts it bluntly: 79% of young Americans do not trust businesses to use AI responsibly. Your employees are part of that public, and they bring those views to work.

Is this just a Gen Z attitude?

No, but younger workers show it most clearly, and they are the heaviest users. Gallup's April 2026 study of Gen Z found weekly-or-more AI use holding steady at 51% while excitement fell from 36% to 22%. The share who agree AI helps them complete work faster dropped 10 points to 56%. Even among daily users, excitement was down 18 points from the year before.

The same study found that among Gen Z workers, 48% now say AI's risks in the workforce outweigh its benefits, up from 37% in 2025, and 69% trust work completed without AI more than work completed with it. These are not people who have never tried the tools. They are people who use them regularly and have formed an opinion.

Why Doesn't Using AI More Often Build Confidence in It?

Because unguided use teaches people what AI gets wrong long before it teaches them how to work with it well. A confident but incorrect summary, an invented citation, or a draft that missed the point all leave an impression. Without a framework for why those errors happen and how to catch them, the lesson employees take away is general: this thing is not reliable.

We want to be careful here. The surveys show skepticism rising alongside use; they do not prove that use causes it. But the pattern is consistent with what we see in client environments, where the teams that were handed access with no structured orientation are the ones most likely to describe AI as hit-or-miss.

The Paradox: Skeptical in Surveys, Uncritical at the Desk

Here is the part that should worry an owner more than the attitude data. The same workforce that tells pollsters it distrusts AI often uses it without checking its work. The 2025 University of Melbourne and KPMG global study of more than 48,000 people across 47 countries found that only 46% of people are willing to trust AI systems. Yet 66% of employees who use AI at work rely on its output without evaluating accuracy, 56% say they have made mistakes in their work because of AI, and 57% say they hide their AI use and present AI-generated work as their own. Only 47% said they had received any AI training.

Put those numbers side by side and the real issue comes into focus. Your team does not have a trust problem in the simple sense. It has a calibration problem. People distrust AI in the abstract and over-trust it in the moment, because nobody has taught them where the line sits for their specific work. We explored the habit-level version of this in our AI Lab piece on the hidden habit undermining your team's AI skills.

The table below shows the three states we see most often, and why only one of them is safe to scale.

Trust stateWhat it looks like at workBusiness consequence
Blind trustOutput is forwarded or filed without reviewErrors reach clients, filings, and financial records
Reflexive distrustTool is avoided or every output is redone by handLicense spend with no productivity return
Calibrated trustEmployees know which tasks need verification and how to do itFaster work with errors caught before they matter

Most teams are not sitting in one row. The same employee can be blindly trusting on a Tuesday afternoon deadline and reflexively distrustful in the next team meeting. That inconsistency is what makes the risk hard to see from the outside.

What Should AI Training Teach to Build Real Trust?

It should teach judgment, not just tool mechanics. A program that shows employees which buttons to click, without explaining where a model is likely to be wrong and which decisions it should never make alone, produces users who become disillusioned the first time the tool fails them visibly. Our main-blog argument that the AI skills gap is not a tool problem makes the same point from the capability side.

Training that moves people toward calibrated trust tends to share a few elements:

  • Failure modes by task. Show employees where AI output typically goes wrong in their own work, such as invented figures in a client summary or outdated terms in a contract draft, so their skepticism has specific targets.
  • Verification methods, not just warnings. "Check the output" is not a method. Teach how: trace a figure to its source, compare against the system of record, or have a second person review anything client-facing.
  • Clear sign-off boundaries. Define which outputs require human approval, especially anything touching customer data, money, or compliance-sensitive work. Our piece on whether human-in-the-loop review is actually working covers how those reviews become rubber stamps.
  • Permission to disclose AI use. When 57% of employees hide their use, mistakes go unreviewed. Make disclosure normal so the right work gets the right scrutiny.
  • Reinforcement after launch. Confidence built in a single onboarding session erodes once people hit edge cases alone. Short follow-ups tied to real examples keep calibration current.

The federal government is moving in the same direction. Our breakdown of the Department of Labor's AI literacy standard shows how evaluation of AI output is becoming a baseline workplace skill rather than an advanced one.

How should leaders measure whether trust is improving?

Not by login counts. Usage tells you people opened the tool, not whether they are using it well. Better signals include how often AI-assisted work is caught with errors in review, whether employees can say which of their tasks need verification, and whether people disclose AI use openly. Those are harder to collect, but they track the thing you care about: whether AI is making your work more reliable or less.

What This Means for the Business

For a 20 to 300 person company, the cost of miscalibrated trust lands in two places. Blind trust shows up as errors in client deliverables, inaccurate figures in reports, and sensitive data handled carelessly, the kind of issues that damage relationships and draw questions from auditors and insurers. Reflexive distrust shows up as paid licenses that do little, and as a workforce that quietly concludes leadership chased a trend. Our main-blog guide to using AI tools without making a mess covers the operational side of that cleanup.

There is also a leadership-credibility cost. Employees who already doubt that businesses use AI responsibly will read a rollout with no explanation and no training as confirmation. Being open about how your company governs AI, not only how employees should use it, is part of earning the trust back. The AI skills every SMB leader should cultivate include exactly that kind of communication.

How Securafy Helps Teams Build Calibrated Trust

When we work with clients on AI skills, we start by finding out where their team actually sits: which tasks people already use AI for, which outputs reach clients or financial records without review, and where people have quietly stopped using the tools. That picture usually shows a mix of over-reliance and avoidance across the same team, and it tells us where training will have the most effect.

From there we build role-based training around the verification habits each function needs, pair it with clear sign-off rules and a disclosure norm, and set up short follow-ups based on real mistakes caught in review. The aim is not to make your team more enthusiastic about AI. It is to make their skepticism precise, so they trust the output that has earned it and check the output that has not.

Where To Go From Here

If your team's confidence in AI is slipping even as usage grows, do not try to talk them out of their skepticism. Give it a method, and you will get both safer work and better results from the tools you already pay for.

If your team is moving faster with AI than your guardrails are, start with structured training rather than another tool. AI University gives your people role-based AI training with security built into the material, not bolted on afterward.

If you would rather talk through your specific environment first, book a strategy call with Securafy and we will walk your current AI usage, exposure, and the fastest path to safe adoption.