Shadow AI is unsanctioned employee use of AI tools, the AI version of shadow IT, and it is a fast-growing source of data leakage. Staff paste confidential data into public chatbots the security team cannot see, creating privacy, IP and compliance exposure. The response that works is not blocking but governance: an acceptable use policy, sanctioned safe tools, visibility, data classification, training, and an AI governance function.
Staff are already using AI, whether an organisation has approved it or not. That is shadow AI, and it is one of the fastest growing risks organisations face, precisely because it is invisible until something goes wrong.
What is shadow AI?
Shadow AI is the use of AI tools and applications by employees without the organisation’s approval or oversight. The classic example is staff pasting work content, a client email, a contract, some code, financial figures, into a public chatbot to draft, summarise or analyse it, without anyone in IT or security knowing.
It is the AI version of shadow IT, the long standing problem of employees using unsanctioned technology outside official channels. Shadow AI has spread faster than shadow IT ever did, for two reasons: useful AI tools are freely available to anyone with a browser, and they deliver immediate productivity gains, so the incentive to use them is strong and the barrier is zero. Employees are usually not being reckless, they are trying to do their jobs better, which is exactly why banning AI outright tends to fail, people route around the ban. Understanding shadow AI as a natural response to useful tools, rather than as misconduct, is the starting point for managing it sensibly.
Why is shadow AI a risk?
The primary risk is data leakage, and its defining feature is that it is ungovernable once it happens. When an employee pastes confidential or personal data into an unsanctioned external AI tool, that data has left the organisation’s control. Depending on the tool, it may be stored, used to train models, or exposed, and the organisation has no visibility of any of it.
That single act cascades into several concrete exposures. It can breach privacy obligations if the data is personal information, expose intellectual property if it is proprietary, and break client contracts if it is client data covered by confidentiality terms. There is also a second class of risk beyond leakage: insecure or hallucinated AI output flowing back into business decisions, code or communications without being checked, because the use is unsanctioned and therefore unreviewed. And because none of it is visible to the security team, the organisation cannot assess or respond to the risk, it simply does not know it is happening. Shadow AI turns a productivity tool into an uncontrolled data channel, which is what makes it a security and compliance problem rather than just an IT preference.
How common is shadow AI?
Shadow AI is widespread, and usage is running well ahead of governance. Industry research consistently indicates that a large share of employees already use AI tools at work, frequently pasting company data into them, often without realising the risk. The gap between how many people use AI and how many organisations govern it is the core of the problem.
The impact is measurable. According to IBM’s 2025 Cost of a Data Breach report, around one in five organisations reported a data breach linked to shadow or unsanctioned AI. Other industry research points to a majority of employees having entered work data into generative AI tools. These figures come from vendor and industry research rather than regulators, so treat the exact numbers as indicative rather than precise, but the direction is unmistakable and consistent across sources: adoption is broad, much of it is ungoverned, and breaches linked to it are already happening. For most organisations, the realistic assumption is that shadow AI is present in their workforce now, whether or not they have detected it.
How do we manage shadow AI?
The most effective response is counterintuitive: make the sanctioned path easier than the shadow path, rather than trying to block AI. Blocking alone tends to fail, because the tools are useful and staff will find ways around a ban, which drives the behaviour further underground and makes it even less visible.
A workable program has several parts. Set a clear acceptable use policy for AI, so staff know what is and is not allowed and why. Provide approved, safe AI tools, an enterprise option with appropriate data protections, so that using AI well is the easy default. Use discovery and monitoring, such as data loss prevention and monitoring of traffic to AI services, to gain visibility of where AI is actually being used. Apply data classification so that sensitive data is identified and kept out of AI tools that should not receive it. Train staff on the risks in plain terms, especially the data leakage point. And establish an AI governance function to own these decisions, which is where a framework like ISO 42001 or the NIST AI Risk Management Framework helps. The combination of a good sanctioned option and real visibility does far more than a policy that only says no.
Where should we start with shadow AI?
Start with visibility and a sanctioned option, in parallel. Nothing can be managed that cannot be seen, so getting some visibility of AI use, through monitoring and simply asking teams what they use, is the first step. At the same time, give people an approved tool that is actually good, because the fastest way to shrink shadow AI is to remove the reason for it.
From there, formalise: an acceptable use policy, data classification guardrails, training, and an owner for AI governance. This connects directly to securing AI systems more broadly, shadow AI is the governance and visibility layer of AI security, the part that comes before any technical control on a model. For an Australian organisation, the privacy dimension makes this urgent: a shadow AI leak of personal information can become a notifiable data breach and a Privacy Act exposure, so the risk is concrete and regulated, not hypothetical. For help setting up AI governance and acceptable use, Cybernion helps through ISO 42001 and broader advisory.
Frequently asked questions
What is shadow AI?
Shadow AI is the use of AI tools and applications by employees without the organisation’s approval or oversight, for example staff pasting work data into public chatbots to draft or summarise content. It is the AI version of shadow IT, unsanctioned technology used outside official channels, and it has spread quickly because useful AI tools are freely available and easy to adopt without anyone in IT or security knowing.
Why is shadow AI a risk?
Shadow AI is risky mainly because of data leakage. Once staff paste confidential or personal data into an unsanctioned external AI tool, the organisation loses control of it, and that can breach privacy obligations, expose intellectual property and break client contracts. It also introduces the risk of insecure or hallucinated AI output flowing into business decisions unchecked. Because it is unsanctioned, none of this is visible to the security team.
How common is shadow AI?
Shadow AI is widespread. Industry research indicates that a large share of employees use AI tools at work, often pasting company data into them, and that a meaningful proportion of organisations have already experienced a data breach linked to unsanctioned AI use. According to IBM’s 2025 Cost of a Data Breach report, around one in five organisations reported a breach involving shadow AI. The trend is that usage is running ahead of governance.
How do we manage shadow AI?
The most effective response is to make the sanctioned path easier than the shadow path. Set an acceptable use policy for AI, provide approved and safe AI tools so staff have a good option, use discovery and monitoring such as data loss prevention to see where AI is being used, apply data classification so sensitive data is not shared, train staff, and establish an AI governance function. Blocking alone tends to fail because staff route around it.
Is shadow AI a privacy risk under Australian law?
Yes. If employees put personal information into an unsanctioned AI tool, the organisation may breach its obligations under the Privacy Act and the Australian Privacy Principles, because it is disclosing or handling personal information without proper control or, potentially, sending it overseas without safeguards. A shadow AI data leak can also become a notifiable data breach. This makes shadow AI a governance and compliance issue, not just an IT one.
Written by Gaurav Vikash, an ASD endorsed IRAP assessor and senior cyber security leader with 18 years of experience across Australia, the UK and Asia, including CISO and senior security leadership roles. He holds CISSP, CISA, CISM and CRISC and is an ISO 27001 Lead Auditor and ISO 42001 Lead Implementer, and speaks regularly at industry conferences.
Sources:
- Cost of a Data Breach Report 2025, IBM, 2025
- Engaging with Artificial Intelligence, Australian Signals Directorate, 2024
- Notifiable Data Breaches, Office of the Australian Information Commissioner, 2026
- ISO/IEC 42001:2023, International Organization for Standardization, 2023
Last updated: 30 August, 2026
