Businesses are hitting pause on AI, but their employees aren’t.

When a business slows down its AI plans, it doesn’t mean the demand for AI inside the organisation slows with it. Employees are still presented with deadlines, still have meeting notes to write up, and still know that a free chatbot that can cut down their workload considerably is just a browser tab away. New research from Acora suggests there could be a gap between what leadership has approved and what staff are actually using, and that gap is where a growing share of AI risk now sits. A pause in activity could be the best moment to close that gap, because you can’t secure what you can’t see.

Acora has analysed four years of UK Google search data across 50 terms that point to unsanctioned AI use, from people asking whether ChatGPT is safe for confidential information, to those looking for ways to delete their chat history or get around a block. Excluding AI “humaniser” tools (which skew heavily towards students), these terms were searched 161,934 times in the UK in the 12 months to September 2026. That is 48% more than two years earlier, and more than two and a half times the volume seen in ChatGPT’s first year.

But first, why are businesses slowing down AI adoption?

FTI Consulting’s AI’s Second Act report, a survey of 1,600 senior decision-makers at large companies, found that 60% have slowed, paused or pulled back a planned AI deployment because of reputational, regulatory or trust concerns. In the UK, that figure was 58%. Data privacy and security breaches were the most commonly cited AI risk, named by 60% of respondents overall and 72% in the UK, while 47% expect to increase investment in cyber security.

That caution is reasonable; most boards would rather get governance right than explain a data leak once things have already gone wrong – the problem is what the pause might look like from an employee’s desk. Under a pause or slowdown of AI investment, an AI tool might not have been sanctioned, but the workload hasn’t changed, and the tempting free alternatives haven’t gone anywhere either. Many employees may decide the risk of using a free, unapproved chatbot is worth taking, inadvertently putting the business at risk.

FTI’s research also found that 54% of decision-makers consider employee use of unapproved “shadow AI” tools a significant concern. Acora’s search data suggests that concern is well founded, and it builds on our earlier analysis of rising shadow AI incidents and the AI use most organisations can’t see.

“AI transformation has reached the point where businesses are asking harder questions about what they can trust, how their data is being used and whether the right controls are in place.

Concerns around trust, governance, security and data shouldn’t leave organisations stuck between ambition and action. We help businesses close those gaps, putting the right data foundations, governance and controls in place so they can adopt AI with confidence, rather than putting promising projects on pause”

David Rabson, Acora CEO

What does the UK search data reveal about shadow AI?

Although search volumes only show us indicative behaviours, the data is candid; people type into Google the questions they probably wouldn’t ask their IT team. Three types of behaviour stood out in the results for how quickly they’ve grown over the past few years:

Together, these three behaviours accounted for more than 100,000 searches in the past year, over three times the level of two years ago.

Are employees worried about putting company data into AI tools?

Yes, and increasingly about specific workplace tools. Searches for “is Copilot safe for confidential information” rose from 73 to 1,558 in a year (an increase of 2,034%). The term “Is Copilot safe to use at work” grew by 159%, and “does Copilot store your data” doubled. Over the same period, searches for “AI data privacy” rose 76%, “ChatGPT data breach” 78% and “does ChatGPT sell your data” 112%.

Most of these searches come from someone who is about to paste something sensitive into a chatbot and wants reassurance first. Some will find a clear company policy, but many will find a blog post or article and make the call themselves.

Copilot is often the approved tool, which reinforces the point that even sanctioned AI can leave employees unsure about what they can share. FTI found that 51% of decision-makers see a lack of employee AI skills or governance as a key risk. Acora’s position is simple: approving a tool is not the same as governing it. A sanctioned tool still needs clear rules on what data goes into it, and people who know how to follow them.

How are employees covering their tracks?

UK searches about deleting or hiding AI chat history reached 10,958 in the past year, almost six times the level two years earlier. “How to delete ChatGPT history” alone was searched 4,052 times, more than five times as often as two years ago, and “can my employer see my ChatGPT history” rose 18% in the last year.

Most of this isn’t sinister; clearing history is a normal privacy habit. But for a business, the result is the same: less record of what data went into which tool, and when. Under UK GDPR and the EU AI Act, being able to show oversight is part of compliance. That’s why Acora treats AI governance as something a business has to be able to evidence, not just write down.

Are people hiding that they’ve used AI at all?

Searches for AI “humaniser” tools, which rewrite AI-generated text so it reads as if a person wrote it, reached 476,119 in the UK in the past year, up 94% on two years earlier. “AI humanizer” alone was searched 320,370 times, more than seven times the level of two years ago.

A large share of this demand is likely to come from students, but the same tools are a click away for anyone who would rather their manager or client didn’t know a report started life in ChatGPT. Where AI use is hidden in the finished work, it can’t be reviewed, checked for accuracy or governed. For Acora, this is one of the clearest signs that AI governance can’t rely on people volunteering what they’ve used.

Why don’t AI blocks work on their own?

Searches for ways to get round AI blocks reached 49,640 in the past year, up 166% from two years earlier, and “unblocked AI” alone grew 24% in the last 12 months. Although many of these searches again most likely come from students, it could very well give us an insight into the people who will be entering the UK workforce over the next few years, who have learned to treat a block as a puzzle, rather than a warning.

Where a block does hold, there are plenty of alternatives. Searches for “free AI no sign up” rose 36% in a year and “AI no sign up” was up 39%. No account means no admin console, no retention settings and no audit trail. Free AI note takers are growing too, with 4,191 searches in the past year, up 28% on the year before and around six times the level of two years ago.

These are tools that sit in meetings and record client conversations, and if one is running on a free account, a transcript of an important board discussion, pricing negotiation or HR case could be stored on servers the business has never vetted, with no easy way to retrieve or delete it. That’s a data protection problem under UK GDPR, and a difficult conversation if a client ever asks where their information went.

How Acora helps businesses use the pause

For the businesses who are slowing their AI activities, the pause doesn’t have to mean standing still. If an organisation has stepped back from a rollout, it has time to understand how its people are already using AI, and to build governance around what’s real. Here’s where Acora can help, with a few steps:

Find out what’s already in use. Look at network and endpoint traffic, browser extensions and the AI features now embedded in existing software. A short, time-boxed discovery exercise usually turns up more than expected. Acora’s data and AI governance team helps organisations build that first inventory, so governance starts from fact rather than assumption.

• Answer the questions your people are Googling. If staff are searching whether a tool is safe for confidential data, the business might not have told them. Only 30% of FTI’s respondents expect to invest more in employee training. Plain-English guidance on what data can go where is one of the cheapest controls available. Acora helps businesses build AI policy from real usage data and turn it into guidance employees can follow with confidence.
• Give people a sanctioned option that’s good enough. Free tools fill the gap when the approved route is slow or missing. A sanctioned tool that does the job removes much of the reason to look elsewhere. For many organisations, that means making sure Microsoft 365 Copilot can only reach the data it should.
• Treat governance as ongoing. New AI tools and features appear every month, so the policy and the discovery work behind it need a regular review cycle. Acora helps businesses build that review into their regular governance, so both the inventory and the policy stay up to date.

Whether a business has paused, slowed down or pressed ahead with AI, the starting point is the same. Acora’s data and AI governance services give organisations clear visibility of the AI already in use, policy built on real usage, and employees who feel confident using AI responsibly.

Whatever stage your AI plans are at, Acora can help you understand how your business is already using AI and build governance around it. Get in touch with our team to start the conversation.