Shadow AI is any AI tool employees use for work without their organization’s approval. It can be a chatbot drafting an email or a transcription tool summarizing meeting notes. When time and resources are tight, using an AI tool seems like an easy shortcut.
But that shortcut can lead to all sorts of problems. At one community bank, an employee uploaded customers’ names, Social Security numbers, and dates of birth into an unauthorized AI tool. The bank’s parent company filed a Form 8-K, and the incident likely set off customer notices and regulatory deadlines. No hacker was involved.
That’s the risk of shadow AI. A well-meaning employee can feed private customer data into generative AI tools, breaching confidentiality and creating a compliance nightmare.
Shadow AI almost never starts with bad intent. It starts with an employee trying to find the tools they need to do their job, so assume positive intent. That assumption shapes everything that follows, from how you detect shadow AI to how you respond to it.
Learn more about the risks of shadow AI, how to detect it, and how enterprise risk management software can support your AI governance program.
Watch the Webinar: Choosing an AI Risk Framework: What Financial Institutions Need to Know
Why Is Shadow AI So Hard to Detect?
Shadow AI is hard to detect because it runs through personal accounts and consumer tools that don't go through the procurement process. Verizon’s 2026 Data Breach Investigations Report found that 45% of employees regularly use AI on corporate devices, and 67% of those users reach the tools through personal accounts outside the organization’s oversight. And you can't govern or disclose AI usage you don’t see.
Related: Using AI in Financial Services: Best Practices and Red Flags
What Risks Does Shadow AI Create?
A single shadow AI incident can have ripple effects across an FI and touch multiple risk areas, including compliance, operations, and data privacy and security.
Shadow AI is a significant source of data breaches, with 20% of organizations reporting a breach related to its use, according to IBM’s 2025 Cost of a Data Breach report. IBM found customer personally identifiable information was compromised in 65% of breaches involving shadow AI, compared to 53% across all breaches studied.
Even if your FI has a written AI policy and approved tools, employees may still turn to unauthorized ones. The Pennsylvania bank offered an approved AI tool, yet an employee still used a personal account on an unapproved tool.
Which compliance obligations apply to a Shadow AI breach depends on your charter and on whether the incident meets each rule’s trigger. Some examples:
- Banks. The Pennsylvania bank’s parent company filed a Form 8-K under Item 1.05, and the incident likely also triggered a 36-hour notice to the bank’s prudential regulator and customer notice under interagency guidance built on the Gramm-Leach-Bliley Act (GLBA).
- Credit unions. The NCUA requires notice within 72 hours of a credit union reasonably believing it has had a reportable cyber incident.
- Mortgage lenders and other non-banks. The FTC’s Safeguards Rule requires covered non-banks to notify the FTC as soon as possible, and no later than 30 days after discovering that 500 or more consumers’ unencrypted information was acquired without authorization.
- RIAs and wealth management firms. SEC-registered advisers must notify affected customers as soon as practicable, and no later than 30 days after becoming aware of unauthorized access to or use of customer information, according to Regulation S-P requirements.
State breach notification laws add their own requirements. As of October 2026, 20 states set numeric deadlines of 30 to 60 days, and the rest require notice without unreasonable delay. Most states also require notice to the attorney general or another state agency. Incidents involving Social Security numbers can also draw class action interest.
Related: What Is Compliance Risk? A Guide for Financial Organizations
Make It Safe to Ask
The most durable defense against shadow AI is an environment where employees bring their tool needs to IT and security before they go around you. If asking feels risky, or the answer is always no, those tools stay in the shadows.
Give people a simple way to request a tool, respond quickly, and treat disclosure as good faith, not an admission of wrongdoing. Detection then becomes a backstop for what slips through, not your first line of defense.
How to Detect Shadow AI
It’s not always easy to find the AI tools employees are using, but it’s not impossible. Here are several tips for uncovering shadow AI:
- Audit corporate devices for browser extensions and AI note-takers that don’t have IT approval.
- Monitor data-loss-prevention alerts as well as network, web proxy, and cloud access security broker (CASB) logs for sensitive data heading to AI platforms and for consumer AI sites employees reach on corporate networks.
- Partner with your accounts payable or purchasing team to review company card purchases and expense reports for individual AI licenses. It is often one of the fastest ways to see where employees are buying productivity tools on their own, and what you find is a map of unmet needs, not a list of violations.
- Survey employees, and consider a no-penalty window for disclosing tools already in use, since logs can miss use on personal devices.
- Check release notes for software employees already use, since an update can add AI features.
As shadow AI and other AI risks continue to evolve, continuous employee training is crucial. Provide training on approved tools and teach team members how to spot AI features in everyday software. Add what you find to your AI inventory, documenting which AI tools employees use and through which accounts.
Managing Shadow AI
Once you identify shadow AI tools employees are using, you have to decide what to do about it.
- Talk to the users. Find out what job the tool does for them. Find out what data they put into it and why they need the tool.
- Assess data exposure. Assess what data is going in (customer info, contracts, code, strategy) to the tool.
- Check for an incident. If regulated, customer, or proprietary data went into a consumer tool, treat it as a possible incident. Assess notification obligations, request deletion where you can, and document what you did.
- Sort each tool into one of two buckets. Approve or block. You may want to find alternatives for blocked tools. If no alternative is provided, employees may continue to use similar tools.
- Put the approved tools through normal vendor due diligence. Review data handling, retention, subprocessors, security attestations, and contract terms, the same as any other third party.
- Report up. Give leadership a one-page view of what you found, the risk, and what changed.
This is where your enterprise risk management software, including your third-party risk management program, helps you. Having an AI governance framework, AI audits, and other AI risk management documents gives you a foundation for making these decisions.
Related: Employee Security Awareness Training Best Practices for FIs
Bring Shadow AI Out of the Shadows
Your employees are already using AI. The question is whether they feel safe telling you. Assume they are trying to do their jobs better, make it easy to ask, and use the signals you already have, like card spend, logs, and surveys, to see the rest. Visibility follows trust, and getting it now costs far less than finding out after an incident.
To see what other risks hide behind a “free” AI answer, download our infographic, The Hidden Cost of Free AI.
