Companies are buying and rolling out more AI than at any point in the past, and a growing body of survey data suggests a large slice of their staff is not using it the way leadership planned. A Ricoh report covered by TechRadar found that 91% of business decision-makers say their AI investments are delivering the expected value, while 45% of workers admit they use non-approved tools, the practice known as shadow AI, at least once a month.
That gap between the boardroom scorecard and the desk-level reality is the story of AI adoption at work in October 2026. The same pattern shows up in Deloitte, Gallup and a separate survey reported by Phys.org. Each measures something different, yet they point the same way: access to AI is spreading faster than trust, training and workflow design. We covered a related control problem earlier this month in our look at how five big US banks block third-party AI agents, and the workplace version of that tension is now playing out inside ordinary companies.
One caveat belongs up front. The available research confirms the gap, but it does not prove that workers are “increasingly” working around company AI over time. None of the surveys cited here publishes a trend line for shadow AI. This article reports what the data does show, flags what it does not, and then analyzes what the numbers likely mean for employers, vendors and workers.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
The Numbers Behind the Workplace AI Adoption Gap
Start with the raw figures, because they carry the argument better than any adjective. Four separate sources, covering executives, workers, consultancies and pollsters, produced the data points below. Read together, they describe an enterprise that has opened the AI tap wide and is now watching the water go in several directions at once.
| Source | Metric | Figure |
|---|---|---|
| Ricoh report | Decision-makers who say AI investments deliver expected value | 91% |
| Ricoh report | Workers who say AI tools create friction or extra work | 53% |
| Ricoh report | Workers using non-approved “shadow AI” tools at least monthly | 45% |
| Deloitte, 2026 State of AI in the Enterprise | Growth in worker access to AI tools in one year | 50% |
| Deloitte, 2026 State of AI in the Enterprise | Organizations that have not redesigned jobs or workflows around AI | 84% |
| Gallup (May 2026) | U.S. employees who say their organization communicated a clear AI-integration plan | 25% |
| Phys.org-reported survey | AI users who say their most-used tool was officially introduced by the employer | 55% |
The headline contrast is the 91% against the 53%. Nearly all the people who sign the checks say the money is working. A majority of the people who touch the tools every day say those tools add friction. Both groups can be telling the truth, because they are measuring different things. Executives look at deployment, licenses and pilots. Workers look at whether the thing saves them ten minutes on a Tuesday afternoon.
What the Ricoh Report Found About Shadow AI
TechRadar’s coverage of the Ricoh report puts three numbers side by side. Ninety-one percent of business decision-makers say AI investments are paying off. Fifty-three percent of workers say the tools create friction or additional work. And 45% of workers say they use unapproved tools at least monthly.
The monthly threshold matters. This is not a count of people who tried a chatbot once out of curiosity. A monthly habit means the unapproved tool has found a repeating slot in someone’s job. When nearly half of workers fall into that bucket, shadow AI stops being a fringe compliance problem and becomes a mainstream behavior that policy has to account for.
It also tells us something about the approved tools. Employees rarely go hunting for alternatives when the sanctioned option works. They go hunting when the official tool is slow, locked down, badly integrated or missing the capability they need. The 53% friction figure and the 45% shadow AI figure likely feed each other, though the report excerpts we reviewed do not prove that causal link, so treat it as the most plausible reading rather than a finding.
Deloitte: Access Up 50%, Daily Use Still Under 60%
Deloitte’s 2026 State of AI in the Enterprise report adds the supply-side view. Worker access to AI tools expanded by 50% in a single year. That is a steep climb for any enterprise technology, and it shows that IT and procurement teams have done what they were asked to do.
Access is not use, though. Deloitte found that fewer than 60% of workers who have AI access use it in their daily workflow. So even among staff who already hold a license or a login, more than four in ten do not fold the tool into a normal day. For a finance chief who approved the spend on the assumption that seats would turn into habits, that is an uncomfortable number.
The more revealing Deloitte figure is that 84% of organizations have not redesigned jobs or workflows around AI. Put differently, only about 16% have (a simple inversion of Deloitte’s figure). Most companies handed employees a new tool and left the underlying process untouched. A chatbot bolted onto a seven-step approval chain does not shorten the chain. It adds an eighth step where the worker copies output from one window into another.
Deloitte also reported that 78% of employees in its survey said they are continually experimenting with new ways of working. That detail cuts against the lazy story that workers resist AI. Experimentation is high. The question is where the experiments happen, inside sanctioned channels or outside them.
Why Workers Route Around Approved AI Tools
No single survey explains worker behavior on its own, but the combined data supports a handful of working explanations. We treat these as analysis, not as findings from the sources.
Friction Beats Policy
When more than half of workers say AI creates friction or extra work, a rational employee does the math. If the approved tool costs more time than it saves, and a free consumer tool saves time, the consumer tool wins. Policy memos rarely beat a faster path to a finished task. This is the same dynamic that has pushed workers toward unapproved tools in earlier waves of software, only now the tools are more capable and easier to reach from a personal phone.
Silence From the Top
Where leadership has not explained what is allowed, workers fill the vacuum with their own judgment. Gallup’s data, covered in the next section, shows how common that vacuum is. An employee who does not know whether pasting a client email into a chatbot is permitted will often decide the answer is probably fine and move on.
Job insecurity adds pressure. Accenture CEO Julie Sweet said in September 2025 that the company planned to cut employees who could not “retrain and retool” on AI. When a CEO of a major consultancy frames AI skill as a condition of keeping a job, workers have an incentive to build that skill by any route available, approved or not. For a broader look at how automation anxiety falls unevenly across the workforce, see our piece on why women face a bigger AI job risk.
Gallup: Only 25% Heard a Clear AI Plan
Gallup found that just 25% of U.S. employees said in May 2026 that their organization had communicated a clear AI-integration plan. Three in four workers, in other words, are either unsure of the plan or have not heard one.
The same Gallup series shows integration itself is moving. The share of U.S. workers saying their organization had integrated AI reached 47% in the second quarter of 2026, up from 41% in the first quarter. That is a six-point jump in one quarter. Companies are clearly deploying. The communication layer is lagging far behind the deployment layer.
| Gallup measure | Period | Figure |
|---|---|---|
| U.S. workers saying their organization has integrated AI | Q1 2026 | 41% |
| U.S. workers saying their organization has integrated AI | Q2 2026 | 47% |
| Change quarter over quarter | Q1 to Q2 2026 | +6 points |
| U.S. employees saying a clear AI-integration plan was communicated | May 2026 | 25% |
Compare the 47% with the 25%. Close to half of workers say AI is integrated where they work, but only a quarter say they know the plan. That mismatch is fertile ground for shadow AI. A worker who sees AI everywhere in the company yet hears nothing about rules has little reason to wait for permission. Agentic tools raise the stakes further, as our coverage of GitHub Copilot controlling desktop apps on two operating systems shows. Software that can act on a user’s behalf is much harder to govern by memo than software that only answers questions.
Phys.org Survey: Usage Barely Moves, Official Tools Lag
A survey reported by Phys.org found that the share of employees using AI at work rose only slightly from the previous year, from 35% to 38%. For a technology that has dominated headlines and capital budgets, a three-point rise is modest.
The same report found that only 55% of AI users said the tool they use most often was officially introduced by their employer. That leaves a sizable minority of AI users leaning on something their company did not roll out, or at least did not say it did. This lines up with the Ricoh finding on monthly shadow AI use, although the two surveys use different samples and wording, so the 45% and the remaining share of the Phys.org respondents should not be treated as the same number.
Taken together, the Phys.org figures complicate the “AI is taking over the office” narrative. Adoption exists and is growing slowly. A good share of it happens outside the official stack. If a chief information officer reports that a certain number of seats are active, that count may miss a parallel population of workers using personal accounts.
How the Three Surveys Compare
The sources differ in who they ask and what they measure. Comparing them helps explain why the headline numbers look so different.
| Source | Who it measures | Core finding | What it implies |
|---|---|---|---|
| Ricoh report (via TechRadar) | Decision-makers and workers | 91% of leaders see value, 45% of workers use shadow AI monthly | Leaders and staff judge success differently |
| Deloitte 2026 State of AI in the Enterprise | Organizations and workers | Access up 50%, under 60% of those with access use it daily, 84% have not redesigned workflows | Tools arrive before process change |
| Gallup (May 2026, Q2 2026) | U.S. employees | 25% heard a clear plan, 47% say AI is integrated | Deployment outruns communication |
| Phys.org-reported survey | Employees using AI | Use rose from 35% to 38%, 55% use an employer-introduced tool | Growth is slow and partly off the books |
No source here is an outlier. The Ricoh data comes from a vendor with an interest in workplace technology, Deloitte sells enterprise advice, and Gallup and the Phys.org-reported survey poll workers directly. When a vendor report, a consultancy and two worker-side surveys land in the same neighborhood, the safest conclusion is that the gap is real even if the exact percentages shift between samples.
Shadow AI vs Shadow IT: The Historical Context
Shadow AI is the latest chapter in a long story. Employees adopted personal cloud storage, consumer messaging apps and their own smartphones years before corporate IT departments sanctioned any of them. In each wave, central teams first tried to block the tools, then discovered that blocking drove usage underground, and finally built approved versions that matched what workers already preferred.
AI differs from earlier waves in two ways. First, the data flows are more sensitive. A worker who pastes a contract clause, a customer record or a snippet of source code into an outside model sends information that may be hard to recall. Second, the tools are far more general. A file-sharing service does one thing. A modern assistant drafts, summarizes, codes and analyzes, so the number of ways it can slip into daily work is much larger.
The software side of this problem is already visible. Our look at AI cutting coding time by 52% while flooding open source with low-quality submissions shows what uneven, unmanaged AI use produces when nobody sets norms. The lesson from earlier technology waves holds: governance that matches how people actually work beats governance that tries to freeze behavior in place.
Risk and Market Impact for Employers and Vendors
For employers, shadow AI creates three overlapping exposures. The first is data leakage, since unapproved tools sit outside logging, retention and contract controls. The second is inconsistent quality, because outputs from a personal tool may never pass the review the company applies to sanctioned systems. The third is measurement. If a company counts only official seats, it will underestimate real AI use and misjudge the return on its spending.
That third point links to the 91% figure from Ricoh. Leaders who feel confident that their investments deliver may be looking at partial data. If a considerable share of useful AI work happens through unapproved tools, official dashboards flatter the sanctioned products on some measures and miss value on others. Either way, the numbers behind the confidence deserve a second look.
What It Means for AI Vendors
Enterprise AI vendors face a product problem, not just a sales problem. The 53% friction number says a majority of workers find current tools cumbersome. Vendors that cut setup time, integrate with the apps people already use and ship admin controls that do not slow users down have an opening. Workplace agent platforms such as those we compared in Claude Duet versus Dots and Muse are competing on exactly this ground, where integration breadth decides whether employees stay inside the approved lane.
Security vendors have their own opportunity. Tools that discover unsanctioned AI use, sandbox it or route it through governed gateways fit the problem the surveys describe. Our walkthrough on how to isolate MCP servers in an AI agent sandbox shows one technical approach teams are using to contain risk without banning the tools outright.
What Is Confirmed and What Is Not
Careful reporting means separating the solid from the speculative. Here is how the claims stack up against the sources.
| Claim | Status | Basis |
|---|---|---|
| Leaders rate AI investments highly while many workers report friction | Confirmed | Ricoh report: 91% vs 53% |
| A large share of workers use unapproved AI tools | Confirmed | Ricoh report: 45% at least monthly |
| Access to AI is expanding quickly | Confirmed | Deloitte: access up 50% in one year |
| Most organizations have not redesigned work around AI | Confirmed | Deloitte: 84% |
| Employees hear little about AI plans | Confirmed | Gallup: 25% in May 2026 |
| Workers are “increasingly” working around AI over time | Unconfirmed | No trend data on shadow AI in the cited excerpts |
The last row is the one that matters for honest framing. The surveys describe a level of shadow AI, not a rate of change. Anyone claiming a surge needs a prior-year shadow AI measurement from the same source, and the excerpts reviewed here do not supply one. The Phys.org figure on overall usage, 35% rising to 38%, is the only year-over-year movement in this set, and it covers all AI use at work, not unapproved use.
Five Predictions for Workplace AI Through 2027
These are our projections based on the data above, not reported facts. We kept them directional and avoided inventing target numbers.
- Workflow redesign becomes the budget line. With 84% of organizations yet to redesign work around AI, consultancies and software vendors will pitch process change rather than more licenses.
- Communication gaps close first. The cheapest fix for Gallup’s 25% figure is a clear internal memo and training plan. Expect many companies to publish AI-use policies before they change any tooling.
- Governed gateways replace blanket bans. Employers that block consumer tools outright will find shadow AI moves to personal devices. Routing use through monitored channels is the more durable answer.
- Measurement shifts from seats to outcomes. Boards that relied on activation counts will ask for time saved and error rates, which will test the 91% confidence level.
- Skill expectations tighten. Accenture’s September 2025 stance on retraining signals a direction other large employers may follow, which will keep pressure on workers to build AI skills on their own time.
What Employers Should Do Next
The data suggests a few practical moves. Ask workers what they actually use, and ask in a way that does not punish honesty. An anonymous survey will tell a company more about its real AI footprint than a license report. Pair that with a plain-language policy that lists approved tools, banned data types and a named contact for questions.
Next, fix the friction. If 53% of workers see extra work, the approved stack has usability problems that policy cannot cure. Shorten onboarding, connect the tools to the systems people already open every morning, and retire pilots that nobody uses. Finally, redesign at least a few high-volume workflows end to end rather than layering AI on top. Companies that do this will have the evidence Deloitte’s figures say most firms lack.
For readers tracking the broader field, our AI and machine learning coverage follows the model launches and enterprise deals that feed these workplace decisions.
Frequently Asked Questions About Shadow AI at Work
What is shadow AI?
Shadow AI is the use of AI tools that an employer has not approved. The Ricoh report, covered by TechRadar, found that 45% of workers use such tools at least monthly.
Are companies getting value from their AI investments?
Leaders say yes. The Ricoh report found that 91% of business decision-makers say their AI investments deliver the expected value. Workers are less convinced, with 53% saying AI tools create friction or additional work.
How many workers with AI access actually use it daily?
Fewer than 60%, according to Deloitte’s 2026 State of AI in the Enterprise report. Worker access to AI tools grew by 50% in one year, so access is expanding faster than daily use.
Do employees know their company’s AI plan?
Most do not. Gallup found that only 25% of U.S. employees said in May 2026 that their organization had communicated a clear AI-integration plan.
Is shadow AI use actually rising?
The sources reviewed here do not say. They show how common shadow AI is today but do not provide a prior-year comparison. The only year-over-year change in this set is overall AI use at work, which rose slightly from 35% to 38%, per a survey reported by Phys.org.
What share of AI users rely on an employer-approved tool?
Only 55% of AI users said the tool they use most often was officially introduced by their employer, according to the Phys.org-reported survey.
Why do workers avoid approved AI tools?
The surveys do not give a single reason. The data points to friction, unclear guidance and workflows that were never redesigned, with 84% of organizations in Deloitte’s data yet to rebuild jobs around AI.
