Glossary
What is shadow AI?
In short
Shadow AI is the use of neural networks and AI services at work without the knowledge or approval of the company, often through personal accounts.
Business and adoption · shadow AI
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine you took your older brother’s tablet to school and told no one. You did not mean any harm; it was just faster for homework. But the teacher does not know what is installed on it, and the tablet still has photos from home. That is why new things at school are agreed in advance, and everyone decides together what can be used in class.
In plain words
An employee wants to finish a report faster and pastes a table into a free chatbot from a personal phone. The security team does not know, there is no contract with the service, and nobody can check where the information went. This is how untracked use of AI appears. Most often people do not break rules on purpose: there are simply no rules, or the company has not given them a convenient approved tool. Personal email and cloud drives went the same way: people used them for work long before they were officially allowed.
For business
This is widespread. In the 2024 Work Trend Index, Microsoft and LinkedIn reported that 78% of AI users bring their own tools to work, and a BCG survey in 2025 found that more than half of employees who lack the AI services they need will find a substitute and use it anyway. For a company this means a risk of leaking personal data and trade secrets, breaking 152-FZ (Russia’s personal data law) and making decisions on unchecked answers. Bans alone help little: people move to personal phones. A working approach has four steps: list what people already use, give company access to a model, write short rules and train staff. Our section on data leaks through AI shows which information leaks most often and how to stop it.
For a specialist
Shadow AI is a special case of shadow IT: the use of models, chatbots, browser extensions and AI features inside approved software without review and approval by the IT and security teams. Typical channels: personal accounts in public chatbots, free AI plugins, built-in assistants of cloud office suites, and self-made integrations through an API. Risks: data goes to a vendor with no contract and no opt-out from training on it, prompts are stored outside the country, there are no logs, and rights to the output are unclear. On 18 August 2025 Fortune, retelling an MIT NANDA report, mentioned the wide spread of shadow AI, meaning unapproved tools such as public chatbots. Detection tools: network traffic and DNS analysis, cloud access security brokers (CASB), data loss prevention (DLP). Ways to reduce it: a register of approved services, company access, an AI use policy and training. In the NIST AI RMF, this task is covered by the inventory of AI systems in the Govern function.
Example
An audit at a trading company showed that managers translated letters to suppliers through personal accounts in three different chatbots. The company connected one company service that does not train on its data and banned personal accounts for work tasks.
Related terms
Questions and answers
Sources
- Microsoft and LinkedIn, 2024 Work Trend Index: AI at work is here, 8 May 2024news.microsoft.com
- BCG, AI at Work 2025: Momentum Builds, but Gaps Remain, 26.06.2025bcg.com
- Fortune (Estrada), MIT report: 95% of generative AI pilots at companies are failing, 18.08.2025fortune.com
- NIST, AI Risk Management Framework, project page (AI RMF 1.0 of 26 January 2023, profile NIST AI 600-1 of 26 July 2024)nist.gov
Not sure whether your company needs this?
AI solutions map