'/%3E%3C/svg%3E){kind=small}
Your CFO installed an AI assistant last week. He gave it access to his email account, SharePoint, and the reporting dashboard. None of this was coordinated with IT.
This is shadow AI. And it’s growing faster than most companies realize.
Shadow AI refers to the use of AI tools and AI agents by employees without the knowledge or approval of the IT department. Similar to shadow IT (unauthorized software running alongside the official IT infrastructure), shadow AI often arises from the desire to work faster.
The key difference: AI agents act autonomously. They don’t just retrieve data. They can perform actions, send emails, create files, and interact with APIs. The scope is fundamentally different.
What distinguishes shadow AI from traditional shadow IT?
Shadow IT was a procurement issue. Someone uses Dropbox instead of the company’s cloud storage. Annoying, but manageable.
Shadow AI is an access issue. An AI agent that an employee has granted access to their work account via OAuth authorization can act on their behalf—create calendar entries, send emails, and open and read documents.
In a finance context, this means: An AI agent with access to a controller’s email inbox can view planning data, forecast comments, and confidential decision-making documents. Whether this is GDPR-compliant often remains unclear. In many companies, no one knows whether the AI provider processes or stores this data.
What specific risks arise for financial data?
Shadow AI risks in finance can be divided into three categories.
Data protection and the GDPR
When financial data is transferred to an external AI provider, a legal basis is required. In the case of consumer AI tools designed for personal use, this basis is usually lacking for corporate data. A GDPR violation is not a theoretical risk; it can have tangible consequences.
Confidentiality and data sharing
AI models are often improved using anonymized user data. What happens to financial data that an employee has fed into a consumer AI agent is often vaguely worded in the terms of service. Anyone who wants to verify this later faces a problem.
Actions with unintended side effects
When an AI agent sends emails or uploads files to file-sharing systems on behalf of an employee, it leaves a trail. During an internal audit or a security incident, it’s difficult to trace the history. There’s simply no log on your end.
Is there shadow AI in your company?
Want to know if shadow AI exists in your organization and what data might be affected? An initial structured check takes less than a day. Let’s look into it together.
How do companies keep track of AI access?
The first reaction of many IT departments: block everything. That doesn’t work. Employees find ways to use AI tools anyway. The result is even less visible shadow AI.
The better approach is visibility plus structure. Three measures can help immediately:
Step 1: Create an inventory
Which AI tools are employees currently using, both officially and unofficially? A quick survey across teams often reveals surprises. Without this inventory, any further action is like groping in the dark.
Step 2: Check OAuth permissions
In most companies, administrators can view the OAuth authorizations for all accounts. A regular audit shows exactly which AI agents currently have access to company data. This is often the most sobering—but also the most valuable—finding.
Step 3: Communicate clear guidelines, don’t just impose them
Employees who use AI tools want to work more efficiently. That’s not a bad motive. A policy that explains why certain tools are problematic for financial data is met with greater acceptance than a blanket ban. Bans without justification only lead to more creative ways to circumvent them.
What does a governance framework for controlled AI use look like?
An AI governance framework for small and medium-sized businesses doesn’t have to be complex. It requires clear answers to a few key questions.
- Who is allowed to use which AI tools for which tasks? Not “everyone” or “no one,” but rather based on role and data sensitivity.
- What data are AI tools allowed to access? Budget data and forecast comments should be treated differently from general market reports.
- How is usage documented? For AI agents operating within the corporate environment, logging should be in place.
- How are new AI tools evaluated and approved? A simple approval process prevents employees from turning directly to consumer-grade versions.
Three BI2run recommendations for a secure start with AI in controlling
Deploying AI agents in controlling and finance isn’t a risk if you approach it in a structured way.
First recommendation: Start with internal data on controlled platforms. Agents running on IBM Planning Analytics or your own cloud infrastructure are, by definition, not shadow AI. You know what they’re doing. You see what they’re processing.
Second recommendation: Use platforms with built-in governance features. IBM watsonx Orchestrate is designed specifically for this purpose. You can view all agents, their access rights, and their activities in one place. No consumer app offers this.
Third recommendation: Treat AI compliance as an ongoing process, not a one-time project. Shadow AI reemerges whenever new tools hit the market. Quarterly reviews help you stay on top of things.
Glossary
Frequently asked questions about shadow AI
How do I know if shadow AI already exists in my company?
A first indication is the OAuth authorization audit: In Microsoft 365 and Google Workspace, admins can view all third-party applications that have access to company accounts. If AI agents appear there that no one has officially approved, that’s a clear sign. A brief anonymous survey of employees can also help.
Is shadow AI automatically a GDPR violation?
Not automatically, but the risk is real. If an AI agent processes personal data or financial data and the provider does not offer a sufficient data processing agreement, there is a compliance issue. Consumer AI tools are often explicitly not intended for processing corporate data.
How can you prevent shadow AI without frustrating employees?
By taking their needs seriously. Employees use shadow AI because they want to work more efficiently. The solution isn’t to ban it, but to provide secure and approved alternatives. Offering employees good tools reduces their motivation to go their own way.
What role does IBM watsonx Orchestrate play in curbing shadow AI?
Orchestrate gives IT departments visibility into all active AI agents and their access rights. This makes shadow AI detectable. In addition, the platform offers a secure, controlled environment for AI agents that is attractive enough to serve as an alternative to consumer solutions.
When should you seek external support for AI governance?
As soon as AI agents begin actively accessing sensitive data, a structured governance framework is advisable. External support is particularly worthwhile if the internal expertise to properly configure IBM platforms is lacking, or if an audit has revealed that shadow AI already exists.
