IAPQB

Staff who bring their own AI

How often staff use AI tools their employer has not approved, what has gone wrong, and how organisations respond

IAPQB Report No. 21 · Practice report · Published 17 November 2026
International AI Practice Qualification Board (IAPQB)

In one minute

  • In a 2024 Microsoft and LinkedIn survey of 31,000 desk workers in 31 countries, 75% used AI at work, and 78% of them brought their own AI tools [1].
  • In a 2024 to 2025 study in 47 countries, 70% of employees using AI at work used free public tools, and 48% of employees had put company information into public AI tools [2].
  • The Dutch Data Protection Authority received breach reports after staff entered patient and customer data into chatbots [3]. In an IBM study, one in five of 600 breached organisations reported a breach due to shadow AI, meaning AI use without approval [4].
  • Guidance now asks for approved tools with work accounts, written rules on tools and data, training, and open communication. The UK National Cyber Security Centre (NCSC) expects some unapproved use to continue [3, 5, 6, 7].

1. A familiar situation

A sales coordinator must turn a long file of customer complaints into a short report by the evening. Her company offers no AI tool, and she does not know whether outside tools are allowed. She pastes the file, with customer names and addresses, into the free chatbot she uses at home. Nobody at her company knows that another company may now store it.

2. Why it happens

Many public AI tools are free and familiar [2, 8]. In an October 2025 survey of 2,003 UK employees, commissioned by Microsoft, 71% had used unapproved consumer AI tools at work. 41% said these were the tools they knew from their personal life, and 28% said their company offered no work-approved option [8]. The NCSC writes that policies and guidance "have not always developed at the same pace" as AI use [7]. Use is often hidden: 52% of AI users in the Microsoft and LinkedIn survey were reluctant to admit using AI for their most important tasks [1]. The NCSC warns that such use can expose company or customer data [7].

IAPQB view. In the surveys below, most AI users used tools their employer had not provided, and 56% of employees had used AI at work without knowing whether it was allowed [1, 2]. In both Dutch cases, the breach was reported after an employee had already entered personal data into a chatbot [3]. Employees chose the tool and the data on their own, and employers learned of it later or not at all.

3. Four cases

3.1 Worldwide, 2024: most AI users brought their own tools

For the Microsoft and LinkedIn 2024 Work Trend Index, Edelman Data & Intelligence surveyed 31,000 desk-based knowledge workers in 31 countries in February and March 2024 [1]. 75% used AI at work, and 78% of them brought their own AI tools. The report says staff acted "without guidance or clearance from the top", putting company data at risk. Only 39% of AI users had received AI training from their company [1].

3.2 Forty-seven countries, 2024 to 2025: company data in public tools

A University of Melbourne team, working with KPMG, surveyed 48,340 people in 47 countries from November 2024 to mid-January 2025 [2]. Of employees using AI at work, 70% used free public tools. 48% of employees had uploaded company information into public AI tools. 34% said their organisation had a policy guiding generative AI, 6% a ban and 41% no policy; 19% did not know. Uploads of company data or copyrighted material were most common under a ban (67%) and least common with no policy (33%). The authors wrote that "outright bans may be ineffective" [2].

3.3 The Netherlands, 2024: patient and customer data reported as breaches

On 6 August 2024, the Dutch Data Protection Authority said it had received a number of breach notifications caused by staff sharing personal data with AI chatbots [3]. A general practice worker had entered patients' medical data into a chatbot, contrary to the agreements. A telecom company worker had entered a file that included customer addresses. The authority said staff often use chatbots "on their own initiative and contrary to the agreements made with the employer", and that entering personal data this way is a data breach [3].

3.4 Worldwide, 2024 to 2025: breaches due to shadow AI

IBM's 2025 Cost of a Data Breach Report, carried out by the Ponemon Institute, covered breaches at 600 organisations from March 2024 to February 2025 [4]. One in five reported a breach due to shadow AI, and only 37% had policies to manage AI or detect shadow AI. Where shadow AI use was high, breaches cost USD 670,000 more on average than where it was low or absent [4].

4. How practitioners handle it now

Staff get an approved tool, with work accounts. The Hamburg data protection authority tells organisations to provide organisation-owned chatbot accounts, and says staff should not open accounts with their private data [5, item 3]. The Dutch authority suggests agreeing with the provider that it will not store the data entered [3]. In a late-2024 trial, the UK Government Digital Service gave Microsoft 365 Copilot to 20,000 government employees. Users reported saving an average of 26 minutes a day, and the report found training essential. Perceived security and data concerns reduced the benefits in a minority of cases [9].

A written rule says which tools, uses and data are allowed. The Hamburg checklist asks for documented internal rules on "whether and under what conditions which tools can be used". Without them, it says, unauthorised use must be expected, and the employer may be liable [5, item 1]. The UK AI Playbook says civil servants must not enter official information into public AI tools unless it is published or cleared for publication [6].

Staff receive training on the risks of AI. Since 2 February 2025, the EU AI Act has required organisations that provide or use AI systems to support their staff's AI literacy. Article 4, amended by the Digital Omnibus on AI in July 2026, keeps this duty with no set level, and national authorities enforce it from 2 August 2026 [10, 11]. The European Commission says it covers a company whose staff use ChatGPT to write advertisement text or translate, and that staff "should be informed about the specific risks, for example hallucination" [11].

Some unapproved use is expected, and staff can say what they use. The NCSC says shadow AI "is unlikely to disappear completely" and sets the goal as reducing the risk. With open communication, it says, staff are "much less likely to turn to shadow IT services, including shadow AI" [7].

Some organisations audit for unapproved AI, and leaks must often be reported. In IBM's 2025 study, 63% of breached organisations had no AI governance policy or were still writing one. Only 34% of those with a policy audited regularly for unapproved AI [4]. If staff leak personal data through a chatbot, the Dutch authority says that telling it and those affected is mandatory in many cases [3].

5. What to take from this

For people who use AI at work. Before work data goes into an AI tool, check that the tool is approved. In the Dutch cases, patient and customer data entered into a chatbot became a reported data breach [3].

For managers. In the Melbourne study, 60% of employees reported no policy or did not know of one, and a ban did not stop uploads [2]. Section 4 pairs an approved tool with a written rule, training and open communication [5, 6, 7, 10].

For schools, training providers and employers. The sources above suggest four training topics: approved tools, data never to be entered, what to do after a mistake, and checking AI output [3, 5, 6]. In the EU, measures for staff AI literacy are already a legal duty [10].

6. Limits of this report

The survey figures are self-reported. Two of the surveys come from Microsoft, which sells AI tools for work [1, 8]. IBM, which sells security software, sponsored and analysed the breach study [4, 12]. KPMG part-funded the Melbourne study, which the university team ran independently [2]. The surveys count different things: own tools, unapproved tools or free public tools. The Microsoft and LinkedIn figure is based on whether the employer provided the tools, so it does not show whether rules were broken [1]. The Dutch authority did not say how many notifications it received [3]. IBM's 2026 announcement gives no new shadow AI figure; the full 2026 report, behind a registration form, was not read [12]. No source measured whether an approved tool reduced unapproved use. Without web search, other cases may have been missed. This report does not test any AI tool.

References

[1] Microsoft and LinkedIn (8 May 2024). AI at Work Is Here. Now Comes the Hard Part. 2024 Work Trend Index Annual Report. Microsoft WorkLab. https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part

[2] Gillespie, N., Lockey, S., Ward, T., Macdade, A., & Hassed, G. (2025). Trust, attitudes and use of artificial intelligence: A global study 2025. The University of Melbourne and KPMG. https://doi.org/10.26188/28822919 (report page: https://kpmg.com/xx/en/our-insights/ai-and-technology/trust-attitudes-and-use-of-ai.html; full report read at https://assets.kpmg.com/content/dam/kpmgsites/xx/pdf/2025/05/trust-attitudes-and-use-of-ai-global-report.pdf)

[3] Autoriteit Persoonsgegevens (Dutch Data Protection Authority) (6 August 2024). Caution: use of AI chatbot may lead to data breaches. https://www.autoriteitpersoonsgegevens.nl/en/current/caution-use-of-ai-chatbot-may-lead-to-data-breaches

[4] IBM (30 July 2025). IBM Report: 13% Of Organizations Reported Breaches Of AI Models Or Applications, 97% Of Which Reported Lacking Proper AI Access Controls [press release on the Cost of a Data Breach Report 2025]. IBM Newsroom. https://newsroom.ibm.com/2025-07-30-ibm-report-13-of-organizations-reported-breaches-of-ai-models-or-applications,-97-of-which-reported-lacking-proper-ai-access-controls

[5] The Hamburg Commissioner for Data Protection and Freedom of Information (13 November 2023). Checklist for the use of LLM-based chatbots. https://datenschutz-hamburg.de/fileadmin/user_upload/HmbBfDI/Datenschutz/Informationen/20231113_Checklist_LLM_Chatbots_EN.pdf

[6] UK Government (10 February 2025). Artificial Intelligence Playbook for the UK Government. https://www.gov.uk/government/publications/ai-playbook-for-the-uk-government/artificial-intelligence-playbook-for-the-uk-government-html

[7] Simon B (7 September 2026). The hidden risks of shadow AI. National Cyber Security Centre (UK). https://www.ncsc.gov.uk/blogs/the-hidden-risks-of-shadow-ai

[8] Microsoft UK (13 October 2025). Rise in ‘Shadow AI’ tools raising security concerns for UK organisations. Microsoft UK Stories. https://ukstories.microsoft.com/features/rise-in-shadow-ai-tools-raising-security-concerns-for-uk/

[9] Government Digital Service (2 June 2025). Microsoft 365 Copilot Experiment: Cross-Government Findings Report. GOV.UK. https://www.gov.uk/government/publications/microsoft-365-copilot-experiment-cross-government-findings-report/microsoft-365-copilot-experiment-cross-government-findings-report-html

[10] European Commission, AI Act Service Desk. Article 4: AI literacy [AI Act text based on the EUR-Lex consolidated version as at 27 July 2026, with the Digital Omnibus on AI amendments marked]. https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-4 (accessed 16 November 2026)

[11] European Commission. AI Literacy - Questions & Answers. Shaping Europe’s digital future. https://digital-strategy.ec.europa.eu/en/faqs/ai-literacy-questions-answers (accessed 16 November 2026; page last updated 20 October 2026)

[12] IBM (29 July 2026). IBM Study: One in Four Malicious Breaches are AI-Enabled, Costing Companies $6 Million on Average [press release on the Cost of a Data Breach Report 2026]. IBM Newsroom. https://newsroom.ibm.com/2026-07-29-ibm-study-one-in-four-malicious-breaches-are-ai-enabled,-costing-companies-6-million-on-average

How to cite

IAPQB (2026). Staff who bring their own AI: how often staff use AI tools their employer has not approved, what has gone wrong, and how organisations respond. IAPQB Report No. 21. International AI Practice Qualification Board. https://iapqb.org/research/021-unapproved-ai-at-work.html

About this report

Prepared by IAPQB. Each fact comes from the sources listed, and each source was opened and checked on 16 November 2026; anything that could not be checked was left out. AI tools were used to search and to draft, and IAPQB checked every claim against its source and is responsible for the text. IAPQB runs an AI work qualification; this report does not promote it. Corrections: contact@iapqb.org or an issue at github.com/iapqb/iapqb.github.io. Licence: CC BY 4.0.

Archived with the series on Zenodo: doi.org/10.5281/zenodo.24248199 · Markdown sources: github.com/iapqb/research