Shadow AI in Mid-Sized Companies: How to Manage Your AI Usage
AI has found its way into mid-sized companies in many ways. However, IT teams haven’t always been involved in this process. While many companies deliberately distribute Copilot or Claude licenses and are aware of additional features in their long-standing CRM or ticketing systems, there is hidden AI in other areas: so-called “shadow AI.” It arises when software vendors fail to disclose their use of AI, or when employees use unauthorized tools for work purposes. This creates significant risks for companies in terms of data protection and security. We explain how companies can ensure transparency and bring AI under controlled use.
Key Takeaways
- Shadow AI encompasses both unauthorized AI use by employees and AI features in existing applications that are not sufficiently known to the IT department.
- AI governance cannot be effectively implemented without transparency regarding where AI is being used. Tools can help with monitoring.
- Mid-sized companies should centrally track the AI tools in use, integrated AI features, data flows, and costs, and clearly define which tool may be used for which purpose.
- For successful AI governance, technical control mechanisms must be combined with an attractive suite of approved AI tools, as well as clear rules and training.
What is Shadow AI?
We primarily refer to Shadow AI in two situations:
1. Unauthorized Use
When employees use AI applications or AI features for business purposes without that use being adequately approved or managed by the company, this is known as “shadow AI.” This could be a personal ChatGPT account or an AI feature that suddenly becomes part of an application already in use. Because AI usage was initially unregulated in many companies, many employees sought out and found ways to use AI: for example, they independently created accounts with OpenAI and fed them company data, or they conducted initial experiments with NotebookLM.
2. Unrecognized Use
In addition, AI functionalities are now present in almost every piece of software or hardware that was already in use: from Breeze at HubSpot to Personio Assistant and Adobe Firefly, all the way to Apple Intelligence on Apple devices. Sometimes the AI is directly visible; other times, it operates hidden in the background. Sometimes it’s active only after consent is given; other times, it works without it. Sometimes you have to allow data training; other times, you have to opt out of it. Sometimes only an account with admin rights can enable features for use; other times, every user can add AI options to their own account.
Why Shadow AI is a Governance Problem for Mid-Sized Companies
When every software application handles AI and transparency differently, new AI features emerge daily, and users employ AI without authorization or oversight, it becomes increasingly difficult for IT teams at mid-sized companies to maintain an overview and ensure AI governance.
They need to know which data is being processed by AI so they can stop the processing if necessary. After all, sensitive data—such as health information—should not be processed by AI. And even if AI features are useful and their use is permissible, IT teams should document this thoroughly and guide their employees in how to use AI.
What should IT teams know about the use of AI in the company?
Among other things, companies should know:
- Which AI tools are being used?
- Which existing applications include AI features?
- What company data is processed by AI?
- Where is data used for training?
- Which personal AI accounts are used for business purposes?
- What is the cost of using AI?
- Are the provided AI solutions actually being used?
These very questions form the basis of effective AI governance.
Reducing Shadow AI: 6 Tips
1. Creating Visibility
Before companies can establish and enforce policies, they must understand which AI is actually being used. While this is relatively easy for officially approved AI services and AI features in other tools, hidden and unauthorized AI is much harder to detect. Tools such as Lookout AI Visibility & Governance or Jamf AI Governance can help with this by identifying where AI is present in mobile apps.
2. Review data flows
It is also crucial to determine where data is transferred and what conditions apply to storage and training.
3. Make and document decisions
Companies should therefore define which tools may be used and which may not. Centrally document approved AI services and AI features integrated into software, and describe how the tools may be used and how data is processed.
4. Formulate clear rules
Employees must be able to quickly understand: Which tool am I allowed to use for what purpose? What data am I allowed to enter? And what is not permitted?
5. Consider costs and usage together
Governance also means cost-effectiveness. Budgets and actual usage should therefore be aligned, and maximum amounts defined for specific time periods.
6. Provide Good Alternatives
If you want to reduce the use of Shadow AI, you shouldn’t just ban it. After all, it’s a good sign that your employees are open to AI and want to improve their work efficiency. So offer them safe solutions for doing so.
Insights into practice
“In our client projects, we first identify which AI tools and features are already in use—from approved AI tools and AI features in existing software to accounts created independently by employees. We combine technical analysis with open dialogue across business departments to uncover previously unknown shadow AI. This overview provides the basis for reviewing data flows and managing AI use effectively.”
Robert Hamel
Solution Architect
EBF
Our conclusion: AI governance requires both technology and people.
Shadow AI is not solely a technical problem. When employees don’t know which tools they’re allowed to use—or when the officially provided solutions don’t meet their needs—workarounds arise almost automatically.
For AI governance to succeed in small and medium-sized businesses, technical control mechanisms must therefore be combined with an attractive selection of approved AI tools, as well as clear rules and training. Only then is productive, controlled use of AI possible.
Contact
Do you have any questions? We’d be happy to talk with you.
Would you like to know where AI is already being used in your company? We can help you create transparency and establish a controlled approach to AI usage.