Shadow AI at Work: Staff Are Already Pasting Company Data Into Free AI Tools

Almost half of employees in a 47 country study by KPMG and the University of Melbourne admit to using AI in ways that break company policy, including uploading sensitive company information into free public tools. And 46% of knowledge workers told Software AG they would keep using their own AI tools even if their employer banned them completely.

What the data shows

• 78% of AI users bring their own AI tools to work, according to the Microsoft and LinkedIn 2024 Work Trend Index.
• 57% of employees say they hide their use of AI and present AI generated work as their own, and only 40% say their workplace has a policy or guidance on generative AI (KPMG and University of Melbourne, April 2025).
• One in five organisations in IBM's 2025 Cost of a Data Breach study reported a breach due to shadow AI, and high levels of shadow AI added USD 670,000 to the average breach cost.
• 97% of organisations that suffered an AI related breach had no AI access controls in place (IBM).
• 39.7% of all AI interactions involve sensitive data, counting prompt text, copy and paste actions and file uploads (Cyberhaven, February 2026).
• 88% of office professionals have shared work information with public AI tools such as ChatGPT, Claude or Gemini (PagerDuty, June 2026).

Shadow AI is the normal state, not the exception

Shadow AI means staff using AI tools the company has not approved, usually a personal ChatGPT, Claude, Gemini or Perplexity account on a work laptop or a phone. It is not a fringe habit.

The 2024 Work Trend Index from Microsoft and LinkedIn found that 75% of knowledge workers already used AI at work, and 78% of those users brought their own tools. Software AG surveyed 6,000 knowledge workers in the US, UK and Germany in September 2024 and found half were shadow AI users. Asked why, 53% said they prefer their independence and 33% said their IT department does not offer the tools they need.

UpGuard's research, reported in November 2025, went further: more than 80% of workers, including nearly 90% of security professionals, use unapproved AI tools, and fewer than half said they knew and understood their company's AI policy. Gartner reported in November 2025 that 69% of cybersecurity leaders have evidence or suspect that employees use public generative AI at work.

For companies in the UAE the starting point is even higher. Microsoft's Global AI Diffusion Report put the UAE first in the world in Q1 2026, with 70.1% of the working age population using AI tools against a global figure of 17.8%. In a 40 or 80 person company in Dubai, the safe assumption is that most of the team already uses AI. The only open question is whose account it runs on.

What employees actually paste into free AI tools

The data that leaves is not random. It is the everyday material of office work.

PagerDuty's June 2026 survey of 1,250 office professionals found that 43% had shared emails or correspondence, 40% meeting notes, 34% customer data, and 31% financial information or confidential company documents and strategies.

Harmonic Security analysed 22,458,240 enterprise AI prompts sent during 2025 and found 579,113 exposures of sensitive data. Code made up 26.5% of what was exposed, legal content 22.3% and merger and acquisition material 12.6%. Of all exposures, 16.9% went through personal free tier accounts, and ChatGPT alone accounted for 71.2% of the total risk.

Cyberhaven's February 2026 analysis shows how much usage sits outside company control: 32.3% of ChatGPT use at work happens through personal accounts, rising to 58.2% for Claude and 60.9% for Perplexity. Sales teams stand out, with sales and go to market data making up nearly 30% of what those teams send to AI.

In a mid sized company this looks ordinary. A sales manager pastes a client list to write a follow up email. An HR lead pastes a performance review to soften the wording. A finance officer uploads a salary sheet to build a chart. Each person is trying to do good work faster. None of them is thinking about where the text is stored afterwards.

The Samsung leak and the ChatGPT search exposure show how it goes wrong

The best known case is still Samsung. In April 2023, engineers in its semiconductor business entered confidential source code and internal meeting notes into ChatGPT while looking for coding help, and MediaNama reported three such incidents within a month. Samsung responded in May 2023 by banning generative AI on company owned computers, tablets and phones and on its internal networks. TechCrunch reported that in an internal survey about 65% of Samsung staff said using generative AI tools carries a security risk.

A second case shows that the risk is not only about training data. In late July 2025, thousands of shared ChatGPT conversations turned up in Google search results, because users had ticked an option to make a shared chat discoverable. OpenAI removed the option, with its chief information security officer saying it gave people too many chances to share things by accident, Engadget reported on 1 August 2025.

The cost is measurable. IBM's 2025 Cost of a Data Breach report put the global average breach at USD 4.44 million. Breaches involving shadow AI exposed more personal information (65% of cases against a 53% average) and more intellectual property (40% against 33%). IBM also found that 63% of breached organisations either had no AI governance policy or were still writing one.

Why banning AI tools at work does not stop the leaks

A ban feels decisive, but the numbers say staff route around it. Software AG's 46% who would keep using personal AI tools under a complete ban is one signal. Salesforce's global survey of more than 14,000 workers found that over half of those using generative AI at work did so without formal approval from their employer, and nearly 7 in 10 had no training on safe use.

Enforcement is patchy too. Among PagerDuty respondents who used AI tools they believed were prohibited, 53% received only informal feedback to stop and 48% faced formal consequences such as a warning. The same survey found 75% would be likely to look for a new job offering better AI skills development. A ban can cost you people as well as productivity.

There is evidence that governance plus approved tools works better. Netskope's 2026 Cloud and Threat Report, covered by Infosecurity Magazine in January 2026, found the share of employees using personal AI accounts at work fell from 78% to 47% in a year, which the magazine read as a sign that data governance policies are starting to bite. The average organisation still logged 223 generative AI data policy violations a month, which is why policy on its own is not the end of the job.

What to do instead: a clear policy, an approved private AI and training

The companies that handle this well follow a similar order.

1. Find out what is already in use. Ask teams which tools they use and for what, and make it clear nobody is in trouble for answering. You cannot govern what you cannot see. Gartner advises regular audits for shadow AI activity, as reported by Infosecurity Magazine.
2. Write a one page AI policy. List what must never go into a public tool: client identity documents, passports, salaries, contracts, source code, anything covered by a confidentiality agreement. In the UAE, personal data processing falls under Federal Decree Law No. 45 of 2021, the Personal Data Protection Law, which applies whether the processing happens inside or outside the country.
3. Give people an approved tool that is better than their personal one. Choose a business plan of Claude, ChatGPT, Microsoft Copilot or Google Gemini, read the terms on data retention and training, and make it the easiest option on every laptop. The 33% who told Software AG that IT does not offer what they need are telling you what fixes the problem.
4. Control who sees what. IBM found 97% of organisations with an AI related breach had no AI access controls. When the company AI can answer questions from company files, each person should only get answers from the files they are already allowed to open.
5. Train in person, on real tasks. KPMG found only 47% of employees had received AI training, and 66% rely on AI output without checking its accuracy. Short sessions built around each team's actual work change behaviour faster than a policy PDF.

Questions people ask

What is shadow AI in simple terms?

Shadow AI is any use of AI at work that the company has not approved or cannot see, most often a personal ChatGPT or Claude account used for work tasks. The risk is that company or client data leaves your control and sits in an account you do not own and cannot audit.

Should we just block ChatGPT on the office network?

Blocking one site does not stop someone using AI on a personal phone, and the surveys above show many staff would keep going regardless. A block works best as part of a plan that also gives people an approved option.

Are paid business versions of AI tools safe for company data?

Business plans generally give the company more control over accounts, retention and administration than personal accounts do, but the terms differ by vendor and plan. Read the data and training terms, switch on the admin controls, and still keep a policy on what never goes in.

Does this matter for a company with only 50 staff?

Yes. Shadow AI comes from individual habits, not company size, and a smaller company usually has fewer people checking where files go. In the UAE, with the world's highest AI adoption rate, a 50 person team very likely has shadow AI already.

CroVIN sets up private enterprise AI on Claude, ChatGPT, Copilot or Gemini for your company and trains your team in person to use it without sending company data where it should not go.