AI Adoption Is a Trust Problem, Not a Tech Problem

Your organization can approve AI tools, but it can't guarantee employees will use them — or even tell you when they don't.

In a 2026 survey, Okta found that 52% of knowledge workers who had used AI in the previous three months said they'd used tools without approval. Meanwhile, 90% of surveyed executives said they were confident they had good visibility into the AI tools their employees were using.

That's a significant gap between what leaders think is happening and what employees say they're doing. Security teams have a name for when employees use unapproved AI tools instead of officially sanctioned ones — "shadow AI" — but the behavior doesn't always come from a desire to break the rules. Sometimes a personal account is just easier to use than the approved alternative.

This comes with downstream problems: Employees may use AI but avoid admitting it because they're worried it will make their work seem less valuable, or raise questions about whether their job could be automated. That can happen even with approved AI tools. Using AI without approval and hiding AI-assisted work aren't the same thing, but both behaviors make it harder for organizations to fully understand how work is changing.

HR can play a role in addressing the conditions behind both behaviors. While IT and security need to make AI safe to use, HR can make AI use rules easier to understand, get managers to model responsible use, and involve employees in decisions that impact their roles and workflows. On The Human Element, HR leaders have described what that looks like in practice. Here's what they had to say.

What Does Shadow AI Cost Organizations?

Employees who turn to unapproved AI tools are often just trying to get their work done, but that still comes with risks.

In PagerDuty’s 2026 survey of office professionals, 88% said they'd shared work-related information with public AI tools like Claude, ChatGPT, or Gemini. 31% said they had shared financial information or confidential company documents.

Unauthorized AI use also creates problems that tend to be less visible, such as:

• Security and governance gaps: If an organization doesn't know which tools employees use or what information they put into them, it can't reliably assess their risks, retention practices, or appropriate access controls.
• Improvements that stay siloed: An employee may discover a genuinely useful AI-assisted process, but if they keep it to themselves, colleagues can't evaluate it, improve it, or incorporate it into supported workflows.
• Misleading adoption signals: An approved platform may appear underused while employees accomplish the same tasks elsewhere. Leaders then have an incomplete picture of what people need from AI and whether their investment is paying off.

There's also a people cost to overly restrictive approaches to AI. In PagerDuty's survey, 77% of respondents said their organization's AI restrictions limited their professional growth or mobility, and 75% said they'd be likely to look for a job offering better AI skills development. That makes unmanaged AI use a retention issue as much as a security risk.

Why Do Employees Go Around Approved AI Tools?

According to the HR experts on The Human Element, there are a few main reasons employees go around approved tools. One is friction; personal accounts are often faster to get to and easier to use. Other reasons are cultural. Many employees don't see their leaders or managers using AI tools, or their workplace culture makes using AI feel like cheating.

Approved AI Tools Take More Effort — or Don't Do the Job

When Okta surveyed knowledge workers who used unapproved AI tools about why they did so, the most common explanations were practical:

• 80% said using their personal accounts was easier
• 78% said their team already used the unapproved tool
• 57% said getting approval for a new tool was too slow or difficult
• 49% said approved tools didn't meet their specific needs

None of these responses are mutually exclusive; an employee can prefer a personal account because it's familiar, because colleagues use it, and because the approved tool lacks a feature they need.

This tells leaders to pay close attention to team habits when investigating unapproved AI use. When a workflow is passed from colleague to colleague, an unauthorized tool can become the normal way of working — often without anyone stopping to question whether it's approved. That's why HR, IT, and team leaders all need a realistic picture of the work employees are trying to accomplish.

Admitting AI Use Can Feel Like Admitting a Shortcut

An employee might be perfectly willing to use AI but reluctant to take credit for using it. In a University of Melbourne and KPMG study, 57% of employees said they hid their AI use and presented AI-generated work as their own.

That statistic points to a workplace question HR can't ignore: What do employees believe will happen if they say AI helped them finish something?

Do they worry that managers will see it as cheating or discount the expertise involved in producing a good result? Are they afraid that if AI can do part of their job, their managers will decide they no longer need a person in the role at all?

Angela Briggs-Paige, Chief People Officer at Acelero Learning, discussed that fear on The Human Element. Her team wanted employees to feel comfortable acknowledging when AI helped draft a communication or summarize a meeting, so leaders made their own AI use more visible and started speaking openly about their AI experiments — whether they worked or not.

"No shame, no secrecy," she said.

Employees Notice When Leaders Follow Different Rules

In PagerDuty’s survey, 81% of respondents believed their leaders operated under different AI rules, even though 86% believed their organizations had AI policies. That perceived double standard makes employees less likely to take any AI policy seriously.

The remedy isn't just another executive endorsement of AI tools or policy. Leaders have to lead by example and show what responsible use looks like: which tools they use, what they use them for, and how they check their outputs. They also need to recognize and address when the approved options aren't meeting employees' needs.

How Can HR Bring AI Use Into the Open?

To bring employee AI use out of the shadows, you need to start by recognizing that access and organizational support aren't competing solutions. 

In BCG's 2026 AI at Work research, 80% of regular AI users who reported strong strategic clarity but limited tool access also reported measurable AI impact. Among respondents with strong tool access but limited strategic clarity, that figure was 60%. Respondents with both strong clarity and strong access reported the highest rate, at 83%.

These are associations in survey responses about AI outcomes — not a direct measure of shadow AI — but they reinforce an important point: Providing access to tools is only part of the job.

Also important to note is that HR can't solve the problem of unsanctioned AI use on its own. Security and IT must establish appropriate technical safeguards, and other departments, like legal, can weigh in on data handling, employment implications, and regulatory requirements. But HR is well placed to learn where employees are struggling, communicate expectations, and help managers respond to these challenges.

Here are five places to start.

1. Remove the Friction from Approved Tools

The difference in convenience is hard to ignore if an employee can open a personal AI account in seconds but has to submit a ticket, wait for approval, and go through training before using an official tool.

That does not mean you should remove security measures or guardrails; you can use tools safely while still giving people a clear, reasonably fast path to access approved tools that can do the work they need done. In partnership with IT and security, HR should strive to answer these questions:

• Can employees easily find which tools are approved, and for which tasks?
• How long does access take, and is the delay proportionate to the risk involved?
• Do approved tools work for the tasks employees are already trying to accomplish?
• Is there a simple way to request a missing capability or propose a new tool?

Zurich Canada shows how this looks in practice; they didn't launch AI tools and leave their employees to sort out the rest. Liz McSavaney said Zurich "took ownership" of the rollout, which meant three things in practice:

1. A safe alternative to public tools: McSavaney described ZuriChat as "a chat GPT with a protective barrier around it that enables employees to use it in a safe way."
2. A clear route to access: "If you want to use Copilot, you have to go through a mandatory training," McSavaney said. Employees are also required to recertify and take a quiz each year.
3. Guidance on best use and safe data handling: The training covers what Copilot and ZuriChat should each be used for, and asks employees to think about "the types of data that they’re putting in," McSavaney said.

The training creates a gate, but it gives employees a clear way through. McSavaney said people have been "really receptive to how we’ve rolled it out."

2. Explain the Reason Behind AI Rules

Your AI policy shouldn't require employees to become experts in data security before they can understand what's allowed. They need enough clarity to make good everyday decisions — and ways to get answers when they encounter situations that don't fit neatly into the guidance.

Diane Sanford, Chief People Officer at Unleashed Brands, emphasized explaining the reasoning behind restrictions her organization put into place; for example, prohibiting AI note-takers in internal meetings because of concerns about proprietary information and the sensitive subjects that could arise.

"A lot of people stop at just creating a framework or a set of conditions or process they want their people to follow. We really believe in explaining the why," she said.

Your AI policy should explain:

• What is permitted. For example, which tools are permitted for drafting, research, summarization, analysis, or other common tasks.
• What information is restricted. What employees must never enter into a tool without appropriate protections and authorization, such as sensitive employee, customer, financial, or proprietary information.
• What needs human judgment. Which decisions, especially those affecting employees, require human review and accountability.
• What to do when something goes wrong. How to report accidental disclosures, questionable outputs, or a tool that doesn't fit the rules.

3. Have Leaders Go First 

Employees take cues from what managers do, not just what your policy says. So a leader who celebrates AI in town hall meetings but doesn't use it themselves sends mixed messages.

Fred Skinner, CHRO at EVERSANA, said executives share their AI wins, openly and loudly.

"Brag about it, because it’s a big deal," he said. "Like, hey, this used to take me four hours. Now I did that in 30 seconds. Take a look."

That peer visibility is important; it allows employees to see approved AI use as an ordinary part of work rather than something to conceal. In the 2026 Work Trend Index, Microsoft found that employees whose managers actively modeled AI use reported a 30-point higher level of trust in agentic AI.

HR can give managers a few habits to practice:

• Mention when an approved AI tool helped with a task, along with the review and judgment the manager contributed
• Invite employees to share responsible experiments, including what they learned when a tool didn't actually help
• Credit the quality of the work and the employee's judgment rather than treating AI assistance as either a secret advantage or a reason to discount the result

4. Talk Early and Openly About How AI Impacts Jobs

Employees' anxiety about AI isn't irrational. Their roles may change, some of their tasks may be automated, and your organization may make workforce decisions as a result. So reassuring everyone that AI will never affect jobs is usually not the right move — and can actively undermine trust.

Briggs-Paige described fear of replacement as a major obstacle to adoption at Acelero.

"If people believe even for a second that AI is coming for their jobs, adoption will fail," she said. "It doesn’t matter how good the tool is. It doesn’t matter how much training you provide."

That's solid advice. Don't leave employees guessing about how AI fits into their role and their future. HR leaders can make those conversations more concrete by explaining what they know and what they don't. For example:

• Which tasks are candidates for automation, and which decisions will remain with people?
• How will performance expectations change when AI makes some work faster?
• What training or opportunities will help employees build skills for redesigned roles?
• What decisions about staffing or job redesign are still undecided, and when will employees learn more?

Finally, be transparent with employees when you don't have the answers yet, rather than making promises you can't keep. Employees are likely to be more understanding of uncertainties, even at the leadership level, than enthusiastic but ultimately unreliable announcements about AI.

5. Let Employees Help Shape AI Rollouts

Employees know what tools they need, when, and where. They also know how to roll out new workflows — which exceptions the workflow contains, which handoffs need human judgment, and which steps look simple on a process map but aren't simple in practice.

Erica Koenig, CHRO at SPS Commerce, has described involving employees in early pilots and letting them help shape AI tools rather than just receiving instructions for a finished rollout. Employees "co-create some of the things that we’re working on," she said, so that "it’s not being done to them, but they’re actually a part of it."

Regardless of the tool you're implementing, you can apply that approach. For example, start with a small group whose work the tool is supposed to improve, and ask them to test real tasks with appropriate data protections, identify where it falls short, and help define what success should look like. Then tell them which changes their feedback produced.

Don't skip that last step; asking for input without acting on it can make participation feel performative. What makes a pilot credible is when employees can see that they have a real say over how the new technology will impact their work.

How Do You Know Whether Trust Is Improving?

While an increase in approved AI usage is an encouraging sign, it doesn't alone prove that employee trust in your organization is improving. Similarly, a reduction in reported shadow AI might mean unauthorized use is on the decline — or it might just mean that employees have gotten more reluctant to admit it, which could indicate trust is actually declining. 

To understand what's changing about employee trust, HR needs to look at both behavior and the conditions behind it. Start with an anonymous survey to establish a baseline, then combine the results with aggregate information from IT and other teams responsible for rolling out and maintaining AI tools.

It can also be worth asking employees directly whether they understand the reason for AI rules and whether they believe that the rules apply evenly to leadership.

It's important to avoid treating any single measure as a trust score. Instead, look for patterns; for example, if employees understand what's allowed, have tools that meet their needs, consistently use approved options, and feel able to discuss the role AI plays in their work, it's likely that trust is strong in your organization.

Make the Approved Path Worth Taking

While shadow AI is often treated as a security failure, it actually shows the demand for AI tools. Employees already think AI is worth using. Now it's up to employers to meet the moment: Create a culture where they don't have to hide, and make approved AI tools the best and easiest to use.

Liz McSavaney described the outcome she's working toward at Zurich Canada: "I would feel comfortable if we had a workforce that was not afraid of AI. They were really clear on how their work was changing and they had far more confidence in using AI to augment their judgment."

If someone on your team found that the approved AI tool couldn't do a job they needed done, would they tell you and help find a solution, or just use their personal account and carry on? The answer tells you something important about both the tools you've provided and the culture surrounding them.