Most owners think their company’s AI use is under control. They bought Copilot or ChatGPT Team, wrote a short policy, and moved on. The 2026 data says otherwise: most AI use at work happens on accounts the company doesn’t own, and a lot of company data goes with it.
This post collects the shadow AI statistics from 2026 reports that are worth quoting, names the source next to each one, and ends with what a small or mid-sized business can do this month. Shadow AI means AI tools used for work that nobody approved, registered or is watching. See what is shadow AI for the short definition.
How much AI use is unapproved
- 67% of people using AI services on corporate devices did so from non-corporate accounts. That’s down only slightly from the year before. (Verizon 2026 Data Breach Investigations Report)
- 45% of employees are now regular users of AI on their corporate devices, approved or not, up from 15% a year earlier. (Verizon 2026 DBIR)
- Shadow AI is now the third most common non-malicious data leakage activity Verizon tracks. (Verizon 2026 DBIR press release)
- 66% of office professionals who use AI said they had used it at work even though they believed company policy didn’t allow it. At organisations with 1,500 or more employees that rose to 72%. (PagerDuty Shadow AI survey, June 2026, run by Wakefield Research with 1,250 office professionals at large companies in the US, UK, Australia and Japan)
- 89% of workers who use AI for their job first started using it outside work. (PagerDuty, June 2026)
The last number explains the first. People learn AI at home, on their own account, and they bring that habit and that account to work. A company licence doesn’t change it on its own.
What goes into those tools
- 88% of the PagerDuty respondents had shared work-related information with public AI tools such as ChatGPT, Claude or Gemini. (PagerDuty, June 2026)
- 34% had entered customer data, and 31% had put in financial information or confidential company documents or plans. (PagerDuty, June 2026)
- 39% said they would rather use AI without telling anyone than disclose it. (PagerDuty, June 2026)
Put the Verizon and PagerDuty numbers side by side and the picture is plain: most AI use runs on personal accounts, and a third of people put customer data into it. Once data goes into a personal account, the company can’t see it, keep it or delete it.
Policy alone isn’t stopping it
- 86% of the PagerDuty respondents believed their company had formal AI policies. (PagerDuty, June 2026)
- 81% believed those rules applied differently to leadership than to everyone else. (PagerDuty, June 2026)
Most companies have a policy, and most staff work around it. A policy matters (see our AI usage policy template), but it only works when someone can see what’s actually in use and act on it.
What it costs when AI is involved in a breach
- About one in five organisations reported a breach involving AI models or applications, and 92% of those lacked proper access controls on their AI. (IBM Cost of a Data Breach Report 2026, 602 organisations, March 2025 to February 2026)
- Only 40% of organisations used access controls on AI models and data. (IBM 2026)
- One in four malicious breaches was AI-enabled, and those breaches cost about $6 million on average, against a global average of $4.99 million. (IBM 2026)
- The FBI counted more than 22,000 complaints involving AI in 2025, with losses above $893 million. (FBI IC3 2025 Annual Report)
The IBM figures are global averages across mostly large organisations, so treat them as a direction, not a small business’s bill. The direction is clear: AI that nobody controls makes breaches more likely and more expensive.
A worked example: a 90-person accounting firm
This is an illustrative example, not a real client.
The firm pays for Microsoft 365 Copilot for its partners and has a one-page AI policy. When it lists what’s actually in use, it finds:
- ChatGPT used by 23 people, 17 of them on personal accounts. Two staff paste client tax summaries in to draft letters.
- An AI note-taker connected to the managing partner’s mailbox and calendar, approved once in 2025. It can read every client meeting.
- A browser extension that summarises web pages and has permission to read every page, including the client portal.
- A Copilot agent built by an office manager that reads a shared folder holding payroll.
None of this breaks the policy on paper. All of it sits outside anyone’s view. That’s the shadow AI gap the statistics describe.
What to do this month
| Step | What it gives you |
|---|---|
| List every AI tool in use: sign-ins, connected apps and browser use | The real picture, not the licence list |
| Give each AI tool an owner and a review date, or remove it | Someone answerable for each one |
| Approve a work account for the tools people actually use | A way to move people off personal accounts |
| Remove AI app access nobody has used in 90 days | Fewer standing doors into mail and files |
| Fix files shared with “everyone” before AI assistants surface them | Less for Copilot or an agent to find |
| Coach people in the moment, not once a year | Better habits without a ban |
Shadow AI detection walks through each place to look, and AI data leaks covers what to do when client data has already gone in.
How Centeye helps
Centeye Lumen is the AI governance and registry platform. It finds which AI tools are in use and by whom across Microsoft 365, Google Workspace and the browser. It shows which AI apps hold access to mail and files and which are riskiest. With the optional browser extension, it shows how much is pasted or uploaded into AI tools, including on personal accounts, without reading the content. Every tool lands in one register with an owner and a review date, and Lumen acts on your rules: coach, block or revoke.
To see where one tenant stands today, run the free AI Exposure Check (or the business version).
Frequently asked questions
What percentage of employees use shadow AI?
Verizon’s 2026 DBIR found that 67% of people using AI services on corporate devices did so from non-corporate accounts. PagerDuty’s June 2026 survey found that 66% of office professionals who use AI had used it at work against what they believed was company policy.
Is shadow AI a security risk or just a policy problem?
Both. Verizon ranks shadow AI as the third most common non-malicious data leakage activity, and IBM found that 92% of organisations with an AI-related breach lacked proper AI access controls. Data pasted into a personal AI account leaves the company’s control.
Does buying an enterprise AI licence fix shadow AI?
It helps, because it gives people an approved option. It doesn’t remove the personal accounts, browser extensions and connected apps already in use. You still need to see what’s there and give each tool an owner.
Where should a small business start?
List every AI tool in use, including connected apps and browser use. Then give each one an owner or remove it, and approve work accounts for the tools people really use.
Figures from the Verizon 2026 Data Breach Investigations Report and its press release; the PagerDuty Shadow AI survey (published 11 June 2026, Wakefield Research, 1,250 office professionals at companies with at least $500 million in revenue in the US, UK, Australia and Japan, 9–20 April 2026); the IBM Cost of a Data Breach Report 2026 (Ponemon Institute, 602 organisations, March 2025 to February 2026); and the FBI IC3 2025 Annual Report. The worked example is illustrative. Microsoft 365 and Copilot are trademarks of the Microsoft group of companies. Google Workspace and Gemini are trademarks of Google LLC. ChatGPT is a trademark of OpenAI. Claude is a trademark of Anthropic.