November 20, 2026

AI Use at Work Jumped 43% Last Year. Most Businesses Still Don’t Know What Their Staff Are Actually Doing with It

AI-Usage

Employees across the UK are turning to AI tools at a pace that most organisations are not equipped to monitor, let alone govern. New data published today by Redflags shows that visits to AI websites on work devices surged 43% year-on-year in 2025, and that the number of companies actively trying to track what their staff are doing with AI nearly doubled in the same period.

The Redflags Behavioural Impact Report 2026 is based on something relatively rare in the research landscape: actual on-device behaviour data. Rather than asking employees how they use AI tools, Redflags measured their behaviour, tracking activity across 44 organisations and delivering over 29 million real-time behavioural nudges throughout the year.

A small group is doing most of the heavy lifting

The standout finding is a concentration of activity that will be familiar to anyone who has studied the Pareto principle in practice. Just 3% of employees account for 18% of all AI-related activity on work devices, generating an average of 235 AI events each, versus 35 for a typical colleague. That’s roughly six to seven times the average usage.

This group is not necessarily doing anything wrong. They may be enthusiastic early adopters who are genuinely productive with these tools. They may also be individuals who are experimenting outside company policy, uploading sensitive files, or using AI platforms without logging in through corporate accounts, any of which can create data exposure risks that are invisible to IT teams without the right monitoring in place.

Tim Ward, CEO of Redflags, said: “Governance is racing to catch up with behaviour, and the gap between the two is where risk lives. The speed at which AI usage is growing inside organisations is remarkable, but what’s equally striking is how many companies are only now starting to understand what’s actually happening on their employees’ devices.”

OpenAI dominates, but the long tail matters

OpenAI accounts for 93% of all AI site visits in the dataset. Gemini comes in at 5%, with Copilot, Perplexity, Claude and DeepSeek each at under 1%. The concentration in a handful of tools might suggest the problem is manageable, but the risk lies less in which tools are being used and more in how they are being used: whether employees are logged into corporate accounts, what data they are sharing, and whether those tools are covered by the organisation’s data processing agreements.

The 9am peak, the most common time for AI site visits, on every working day of the week, suggests AI has become as embedded in the morning work routine as checking email. For business leaders, that normalisation is both an opportunity and a governance challenge.

The human behaviour problem behind the technology risk

Redflags frames AI adoption primarily as a behavioural challenge rather than a purely technical one, and the science backs that framing. The human brain is predisposed to seek novelty – new tools generate dopamine responses that make them inherently compelling, regardless of approval status. Employees are also responding to top-down pressure: with AI positioned by boards and the media as transformative, many feel a need to demonstrate AI fluency simply to protect their professional standing.

Ward noted: “Understanding that this is a human behaviour challenge, not just a technology policy one, is critical to building an effective response. The human brain is wired to seek novelty – new AI tools trigger dopamine responses associated with excitement and reward. This makes them inherently compelling to employees, regardless of whether they’re approved by the business.”

The report argues that blanket bans on AI tools tend to create a compliance gap, policies get ignored, or slow down legitimate work, whereas behavioural nudges delivered at the point of risk can steer employees toward approved tools without disrupting their workflow.

Phishing remains the most persistent baseline risk

The AI findings sit alongside a blunter data point: 93% of organisations in the study were tracking risky behaviour around clicking links in emails from unknown senders. Despite years of awareness training, this remains the most commonly monitored employee risk.

The Redflags data does offer some cause for optimism. Behavioural nudge interventions produced an average 35% reduction in dangerous link clicks, with peak reductions of 83% in the best-performing organisations. Crucially, the hover-to-click ratio, the number of times employees hovered over a link before deciding whether to click, increased by 28% over six months, suggesting that repeated nudging builds more cautious habits over time rather than being ignored after the first encounter.

A 22% average reduction in passwords entered on sites accessed via unknown-sender email links provides further evidence that the effects compound: nudges at the point of email interaction also reduce the likelihood of credential theft at the destination site.

The business case for behavioural security

For UK businesses navigating AI adoption at speed, the Redflags report makes a practical case for investing in visibility before governance. Organisations that can see what tools their employees are actually using, and how, are better placed to write policies that reflect reality rather than assumptions, and to intervene at the moments that matter.

The report covers data from 44 organisations across financial services, engineering and manufacturing, government, and other sectors. The full Redflags Behavioural Impact Report 2026 is available here: https://info.redflags.io/redflags-behaviour-change-report-2026