Understanding Shadow AI: Risks, Costs, and Governance

According to writers at IBM, shadow AI is, “the unsanctioned use of any artificial intelligence (AI) tool or application by employees or end users without the formal approval or oversight of the information technology (IT) department.” 

What Should Special Librarians Know About Shadow AI?

Previously, I wrote about the importance of AI policies in special libraries, and hope your organizations have moved to develop policies or guidelines for the use of AI. When properly implemented, AI-powered tools can greatly enhance search, discovery, and operational efficiency in libraries. However, the rise of “shadow AI” highlights the need for oversight to maximize benefits while managing potential risks. 

Something to be aware of is that “shadow” AI usage can bring financial risks beyond compliance concerns. Some AI platforms employ usage-based pricing models, similar to how libraries pay per reader access for e-books. Without centralized oversight, costs can escalate quickly and unpredictably when employees independently adopt AI tools. 

Common AI Pricing Models

As AI capabilities become increasingly embedded in business software and services, organizations need to carefully evaluate the various pricing structures offered by vendors. Some common models include: 

• Pay-per-token or pay-per-query pricing 

• Monthly subscription tiers based on usage volumes 

• Enterprise licensing with predetermined usage limits 

• Hybrid models combining fixed and variable costs 

Each pricing model has different implications for budgeting and cost control. For example, usage-based pricing offers flexibility but can lead to surprise expenses if usage spikes unexpectedly. Fixed-cost models provide more predictable budgeting but may result in paying for unused capacity. 

Understanding Your Library’s AI Needs

To make an informed decision, organizations should: 

1. Audit current and projected AI usage across departments 
2. Compare total cost of ownership across different pricing models 
3. Consider implementation and integration costs 
4. Evaluate scalability needs as AI adoption grows 
5. Account for potential training and support requirements 

The key is not to automatically reject or embrace any particular pricing model, but rather to thoroughly assess your organization’s specific needs, usage patterns, and budget constraints. This evaluation should be part of a broader AI governance strategy that balances innovation opportunities with cost management and risk control.

By proactively creating an AI policy and governance strategy, libraries and other organizations can harness its power while maintaining cost efficiency and oversight.