Shadow AI statistics 2026: the numbers, with their sources

Shadow AI is now the default, not the exception

In 2026 surveys, 49% of employees (BlackFog, January 2026) and 66% of office professionals (PagerDuty, June 2026) use AI tools their employer has not approved, and IBM reports shadow AI in 43% of breach incidents, up from one in five in 2025.

Shadow AI is the use of AI tools by employees on company work without approval, oversight or audit. If the term is new to you, start with what shadow AI is and why banning it fails. This page collects the figures that measure it, each one linked to the organisation that produced it.

How this page is built

Three rules, so the numbers stay usable:

• Primary sources only. Every figure links to the report or press release of the organisation that ran the survey or the measurement. Where the full report sits behind a registration form, the page says so and names the outlet that reported the figure.
• Date and sample next to the number. A 2024 survey of 31,000 people and a 2026 poll of 1,250 office workers do not measure the same thing. The date and the sample travel with each figure.
• Nothing unverified. Figures that circulate widely but cannot be traced to a primary source are left out.

Last checked: 29 September 2026.

Key figures at a glance

Figure	What it measures	Source	Date
78%	AI users who bring their own AI tools to work	Microsoft and LinkedIn Work Trend Index	May 2024
57%	Employees who hide their use of AI	KPMG and University of Melbourne	April 2025
90% vs 40%	Companies whose workers use personal AI tools, vs companies that bought an official LLM subscription	MIT NANDA, The GenAI Divide	July 2025
USD 670,000	Extra breach cost with high shadow AI	IBM Cost of a Data Breach 2025	July 2025
47%	Generative AI users on personal AI apps	Netskope Cloud and Threat Report 2026	January 2026
49%	Employees using AI tools their employer has not sanctioned	BlackFog	January 2026
39.7%	Data movements into AI tools that involve sensitive data	Cyberhaven 2026 AI Adoption and Risk Report	February 2026
66%	Office professionals who have used unauthorized AI tools	PagerDuty	June 2026
43%	Breach incidents involving shadow AI	IBM Cost of a Data Breach 2026	July 2026

How many employees use unapproved AI

The baseline is high and it has been for two years.

• 75% of knowledge workers use AI at work, and 78% of those AI users bring their own tools. Survey of 31,000 people in 31 countries. Microsoft and LinkedIn, 2024 Work Trend Index, 8 May 2024.
• 58% of employees intentionally use AI at work, and 57% hide their use of AI and present AI-generated work as their own. Almost half admit to using AI in ways that contravene company policy. Over 48,000 people in 47 countries, surveyed November 2024 to January 2025. KPMG and the University of Melbourne, April 2025.
• Workers from over 90% of the companies surveyed reported regular use of personal AI tools for work, while only 40% of companies said they had purchased an official LLM subscription. The authors call it “the shadow AI economy”. MIT NANDA, The GenAI Divide: State of AI in Business 2025, July 2025.
• 49% of employees use AI tools not sanctioned by their employer, and 86% use AI tools at least weekly. 2,000 employees in the UK and US at organisations of more than 500 people, fieldwork November 2025. BlackFog, 27 January 2026.
• 66% of office professionals have used unauthorized AI tools at work. 1,250 office professionals at organisations with at least USD 500 million in revenue, in the US, UK, Australia and Japan, IT roles excluded, conducted by Wakefield Research. PagerDuty, 11 June 2026.

Personal accounts and free tiers

Personal accounts are where company data leaves every enterprise control: no SSO, no retention setting, no audit log. Where an AI tool processes data, and under which law, is the first question of sovereign AI.

• 47% of generative AI users use personal AI apps, down from 78% a year earlier. Users switching back and forth between personal and enterprise accounts grew from 4% to 9%. Netskope Cloud and Threat Report 2026, January 2026.
• 32.3% of ChatGPT usage and 24.9% of Gemini usage goes through personal accounts. Based on observed data movements, not a survey. Cyberhaven 2026 AI Adoption and Risk Report, 5 February 2026.
• 58% of employees using unapproved AI tools rely on free versions. BlackFog, 27 January 2026.

What data goes in

The exposure is ordinary work data, pasted a little at a time.

• 39.7% of all data movements into AI tools involve sensitive data. Cyberhaven, 5 February 2026.
• 88% of office professionals have shared work-related information with public AI tools such as ChatGPT, Claude or Gemini, and 34% have input customer data. PagerDuty, 11 June 2026.
• 33% of employees have shared research or data sets on unsanctioned AI tools, 27% employee data and 23% financial statements or sales data. BlackFog, 27 January 2026.
• Generative AI data policy violations doubled in a year, while the number of generative AI users tripled. In the average organisation, 3% of generative AI users commit an average of 223 data policy violations per month. Netskope, January 2026.

The governance gap

Adoption ran ahead of policy, training and approval paths.

• Only 47% of employees say they have received AI training, and only 40% say their workplace has a policy or guidance on generative AI use. KPMG and the University of Melbourne, April 2025.
• 63% of breached organisations either have no AI governance policy or are still developing one, and only 37% have policies to manage AI or detect shadow AI. IBM Cost of a Data Breach Report 2025, 30 July 2025.
• 69% of cybersecurity leaders have evidence or suspect that employees use public generative AI at work. Gartner, 19 November 2025.
• 50% of organisations lack enforceable data protection policies for generative AI apps. Netskope, January 2026.
• 63% of employees think it is acceptable to use AI tools without IT oversight if no company-approved option is provided, and 51% have connected AI tools to other work systems without IT approval. BlackFog, 27 January 2026.
• 69% of respondents at President or C-level believe speed trumps privacy or security, against 37% in administrative roles. BlackFog, 27 January 2026.

Breaches and cost

This is where the risk shows up on a P&L.

• One in five organisations reported a breach due to shadow AI. Organisations with high levels of shadow AI paid an average of USD 670,000 more per breach than those with low or no shadow AI. 97% of organisations that reported a breach of AI models or applications had no AI access controls in place. IBM Cost of a Data Breach Report 2025, 30 July 2025.
• Shadow AI was involved in 43% of security incidents in 2026, more than double the previous year’s share. The full report sits behind a registration form; the figure is reported by Cybersecurity Dive from the IBM Cost of a Data Breach Report 2026, 29 July 2026.
• The global average cost of a breach reached USD 4.99 million, and one in four malicious breaches was AI-enabled, at USD 6 million on average. 602 organisations breached between March 2025 and February 2026, research by the Ponemon Institute. IBM press release, 29 July 2026.
• By 2030, more than 40% of enterprises will experience security or compliance incidents linked to unauthorized shadow AI. A prediction, not a measurement. Gartner, 19 November 2025.

What the numbers say together

Read side by side, the surveys point in the same direction even where their percentages differ.

Shadow AI is a supply problem before it is a discipline problem. 63% of employees find unsanctioned AI acceptable when no approved option exists, and 40% of companies bought an official subscription while workers at 90% use personal tools. People reach for what works. Where the approved path is missing or weaker, the unapproved one wins.

Personal accounts are shrinking, sensitive data is not. Netskope sees personal app use fall from 78% to 47%, yet Cyberhaven finds sensitive data in 39.7% of data movements into AI tools. Moving people onto enterprise licences does not by itself decide what they paste, which skills they reuse or who can see the result.

The expensive part is the missing trace. IBM’s figures tie the extra cost to the absence of access controls and governance policies. An organisation that cannot say who used which AI capability, on what data, cannot scope a breach quickly. That is why an execution trace (who ran what, on which data) matters as much as a licence. The same logic applies to the skill files employees now share outside any control, covered in why SKILL.md became the king of shadow AI.

For a quick read of where your own organisation stands on capture, control and traceability, the 10-question AI check-up takes two minutes and asks for no email.

Cite this page

skilder, “Shadow AI statistics 2026: the numbers, with their sources”, www.skilder.ai/en/blog/shadow-ai-statistics-2026/, last checked 29 September 2026. Please cite the primary source next to each figure as well.

Frequently asked questions

What percentage of employees use shadow AI? +

It depends on the survey and its definition. In 2026, 49% of 2,000 UK and US employees told BlackFog they use AI tools their employer has not sanctioned (January 2026), and 66% of 1,250 office professionals told PagerDuty they have used unauthorized AI tools at work (June 2026). Earlier, Microsoft and LinkedIn found that 78% of AI users bring their own AI tools to work (May 2024).

How much does shadow AI cost in a data breach? +

IBM's Cost of a Data Breach Report 2025 found that organizations with high levels of shadow AI paid an average of USD 670,000 more per breach than those with low or no shadow AI, and that one in five organizations reported a breach due to shadow AI. The 2026 edition reports shadow AI in 43% of incidents.

Are employees still using personal AI accounts at work? +

Yes, though less than before. Netskope's Cloud and Threat Report 2026 (January 2026) found that 47% of generative AI users use personal AI apps, down from 78% the year before. Cyberhaven's 2026 report (February 2026) found 32.3% of ChatGPT usage and 24.9% of Gemini usage goes through personal accounts.

Can I cite these shadow AI statistics? +

Yes. Cite the primary source linked next to each figure, and this page as the compilation: skilder, Shadow AI statistics 2026, www.skilder.ai/en/blog/shadow-ai-statistics-2026/. Figures come from different surveys with different samples and definitions, so they are not directly comparable with each other.