Why did the AI rollout you paid for stall right after launch, even though nobody pushed back in the kickoff meeting? For a growing number of companies the uncomfortable answer is that some employees are quietly working against it. In April 2026, WRITER and Workplace Intelligence reported that 29% of employees admit to sabotaging their company's AI strategy, a figure that rises to 44% among Gen Z workers. Employees sabotaging an AI rollout is no longer a fringe problem. It is a predictable response to how most rollouts are designed.
The direct answer: employees sabotage AI rollouts when the rollout gives them more reasons to protect themselves than to participate. That usually means unclear intent about jobs, punitive adoption mandates, tools that do not fit real work, and no safe way to raise concerns. Fix those four conditions and most resistance turns into usable feedback.
That framing matters because the instinct of most leadership teams is to treat sabotage as a discipline problem. It is better understood as a design problem, and design problems belong to the people who designed the rollout.
It rarely looks like open refusal. According to Fortune's reporting on the WRITER study, the behaviors employees admitted to include entering proprietary information into public AI tools, using unapproved tools, refusing to use the sanctioned ones, tampering with performance reviews, and deliberately producing low-output work so the technology looks less effective than it is.
Look at that list from an owner's point of view and two very different problems appear. One is a productivity problem: a pilot that is being quietly starved of good inputs will never show the results you budgeted for. The other is a security problem. An employee who routes around the approved tool and pastes client data into a free chatbot has not just resisted your strategy, they have created a data exposure your policies were written to prevent. The same survey found that 67% of executives believe their company has already suffered a data leak or security breach because an employee used an unapproved AI tool.
None of this shows up cleanly on an adoption dashboard. Login counts can look healthy while the work underneath is being slow-walked, which is why leadership teams are often the last to know a rollout is in trouble.
Not by itself, and this is where the common explanation falls short. Job fear is real, but among workers who admitted to sabotage, only 30% cited fear that AI would take their job. Fewer than a third of saboteurs named the reason most headlines lead with.
Gallup's research points the same way. In a February 2026 study of more than 23,000 U.S. employees, Gallup found that among non-users in organizations that offer AI, 46% prefer to keep doing their work the way they do it now, 43% are ethically opposed, 43% have data privacy, security, or compliance concerns, and 39% do not believe AI can help with their work. Only 10% said they were concerned about their role being replaced.
Here is the take most leaders resist: a large share of what looks like sabotage is a vote of no confidence in the rollout itself. People who think the tool is useless for their job, risky for client data, or ethically questionable are not going to become enthusiastic because a deadline is approaching. They are going to comply on paper and work around it in practice. That is a signal about fit, governance, and communication, and it is far cheaper to hear it early than to discover it in a failed renewal.
Early-career roles are concentrated in exactly the tasks AI is marketed to absorb: drafting, summarizing, scheduling, and first-pass research. Those employees can see most clearly what a successful rollout might cost them personally, and the 44% sabotage rate among Gen Z respondents in the WRITER data reflects that exposure. For a 50-person firm, that matters because the junior staff who resist are often the same people you were counting on to carry the day-to-day use.
Because it confirms the fear it is meant to overcome. The same WRITER survey found that 60% of executives plan to lay off employees who can't or won't use AI, 77% say employees who refuse to become AI-proficient won't be considered for promotion, and 69% of C-suite respondents say their company is already doing layoffs due to AI. When employees hear that message, protecting themselves becomes the rational move.
WRITER's own analysis says it plainly: when managers can't guide AI adoption and executives resort to threats rather than transformation, employee resistance becomes the rational response. The same analysis found that only 35% of employees say their manager is an AI champion, and 79% of organizations report challenges adopting AI despite heavy investment. Pressure from the top is filling a gap that should have been filled by explanation and support in the middle.
There is a business cost to the threat approach that rarely gets counted. Mandates produce surface compliance, which inflates adoption metrics, which convinces leadership the rollout is working, which delays the honest conversation until money has already been spent. We see the same pattern in security awareness programs: as our piece on leadership's role in building cyber awareness argues, culture follows what leaders model and reward, not what they threaten.
Treating each form of resistance as information makes the fix more obvious. The table below pairs the behaviors reported in the WRITER research with what they usually indicate and what actually changes them.
| What you see | What it usually signals | What changes it |
|---|---|---|
| Quiet non-use of the approved tool | The tool does not fit the workflow, or nobody showed how it applies to this role | Role-specific use cases and a manager who uses it visibly |
| Use of unapproved AI tools | The approved tool is slower or weaker for real tasks | A usable sanctioned option plus a clear, short use policy |
| Client data pasted into public chatbots | No one explained what is off-limits or why | Data handling rules people can remember, backed by technical controls |
| Deliberately weak output to make AI look bad | Employees believe a successful pilot leads to cuts | Explicit commitments about how productivity gains will be used |
The pattern across every row is that the fix sits with leadership, not with the employee. That is good news, because it means the problem is within your control.
Start by changing what employees stand to lose and gain, then make the safe path the easy path. Our leadership playbook for AI adoption in SMBs covers the broader sequence; for a rollout already under strain, these steps do the most work:
Notice what is missing from that list: a firmer deadline. Deadlines work when the obstacle is procrastination. They fail when the obstacle is distrust, and distrust is what the data describes.
Treat it as its own workstream rather than folding it into the adoption plan. Shadow AI use does not stop because morale improves; it stops when the sanctioned tool is good enough and the guardrails are clear. Our main-blog guide to securing Microsoft Copilot and AI agents before they become shadow IT covers the technical controls, and the broader lesson from why most SMB AI initiatives fail applies here too: tools without governance create risk faster than they create value.
The costs are real but scattered, which is why they rarely get added up. Licenses go unused while renewals come due. Pilots run long because inputs are poor, then get judged as failures. Managers spend time policing usage instead of improving work. Meanwhile the data exposure from unapproved tools accumulates quietly until a client, an auditor, or a cyber insurer asks how AI is being used with sensitive information.
There is also a talent cost. The employees most likely to resist are often early-career staff you will need in three years, and the executives in the WRITER survey are openly building an "AI elite" tier of super-users. A rollout that splits a team into favored adopters and threatened holdouts will cost you people on both sides of that line.
When we work with a client whose AI rollout is underperforming, we start by separating the two problems hiding inside it. On the adoption side, we look at which roles are using the sanctioned tool, which are not, and what those non-users are doing instead, then sit down with managers to find where the tool does not fit the actual workflow. That conversation usually surfaces the real objections faster than any survey.
On the risk side, we check where company data is flowing into unapproved AI tools, tighten the controls in Microsoft 365 and the endpoint stack, and help leadership write a plain-language statement of intent and use policy that employees can trust. The goal is a rollout where the safest way to use AI is also the easiest one, and where employees can see what the business is asking of them and why.
If your AI rollout is meeting quiet resistance, the fix is rarely more pressure. Start by finding out what your people are actually objecting to, then redesign the rollout around those answers before you renew another license.
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.