1 min read
Artificial Intelligence in Banking: Innovation Without Losing Control
Artificial Intelligence (AI) has quickly become a daily business tool. Employees of financial institutions are using AI-powered assistants to draft...
2 min read
Stephanie Goetz : Updated on November 20, 2026
Artificial Intelligence (AI) is wanted, and it’s wanted right now by many Community Bankers. You may be finding yourself wondering about the expectations for governing AI well from a management perspective or what the examiners will be expecting.
The Conference of State Bank Supervisors, CSBS, has released some guidance, and I can tell you it’s much more readable than the previous exam handbooks. Those things were painfully long and vague, and brutal! Anyway, I’ll break it down a piece for you today: AI inventory.
Why do you need an inventory? Well, simply, if we don’t know it exists, we can’t manage it. We can’t manage the risk to our operations, decisions, customers, etc. Very likely this needs to be managed from a third-party risk perspective as well, unless you’ve been developing and training your own AI model…which is highly unlikely.
Any software, system, service, model, tool, or feature that uses machine learning, large language models, predictive analytics, decision optimization, classification, recommendation engines, computer vision, speech recognition, or generative AI.
Include:
Also note that the CSBS wants this inventory to include use cases, not just the tools themselves. So, you will likely have one tool in many use cases.
Example: Instead of ‘Microsoft Copilot’, use:
These will all be related to Copilot; however, each will have their own risk rating.
For each AI use case, capture:
|
Field |
Description |
|
AI Use Case Name |
Fraud Detection, Call Center Copilot, etc. |
|
Business Owner |
Responsible executive |
|
Department |
Lending, Operations, Compliance, Marketing |
|
Technical Owner |
IT owner |
|
System |
Product name |
|
Vendor |
Vendor name |
|
Production Status |
Proposed, Pilot, Production, Retired |
|
Field |
Description |
|
Purpose |
Why AI is used |
|
Business Process Supported |
Lending, BSA, HR, Customer Service, etc. |
|
Field |
Description |
|
Data Sensitivity |
Low, Moderate, High |
|
Consumer Impact |
None, Indirect, Direct |
|
Decision Making Impact |
Informational or Determinative |
|
Regulatory Impact |
Lending, UDAAP, BSA, Fair Lending, etc. |
The CSBS framework includes an AI Use Case Risk Tiering Worksheet to help with the risk classification. See https://www.csbs.org/csbs-artificial-intelligence-supervisory-framework
If you’re not already involved in your Third-Party Risk Management program, they will be a great resource. Most purchased software and services incorporate AI already; it's estimated to be 50-75% of Third Parties.
You may also look at existing process and software inventories in your IT/Information Security Risk Assessment, Business Impact Analysis (BIA), etc. will all be great starting points to identify processes, systems, and their owners.
Once you have those lists, interview those owners. I wouldn’t send surveys, as they likely won’t be experienced enough to fill out a form with the judgments required to complete most of the information.
Utilizing the AI Use Case Risk Tiering Worksheet, determine the overall risk of each use case. Fundamentally, this utilizes data sensitivity and operational impact to calculate that overall risk. Based on that risk level, it will drive the due diligence and oversight required. Check it out at: https://www.csbs.org/csbs-artificial-intelligence-supervisory-framework
Need a Framework for Managing AI Across Your Bank? Explore our AI Risk Management & Oversight Program.
1 min read
Artificial Intelligence (AI) has quickly become a daily business tool. Employees of financial institutions are using AI-powered assistants to draft...
1 min read
What are the regulatory requirements for AI governance for the financial industry?
1 min read
Many financial institutions are asking, “What should our board know about AI?” The answer is practical rather than technical: boards do not need to...