Your Vendors Are Using AI on Your Data - And You Probably Don't Know It

Every week, employees across your organization, and across every vendor organization that touches your data, are using AI tools that your security team never approved, never reviewed, and cannot see. They're pasting customer records into chatbots. Feeding source code to coding assistants. Summarizing confidential deal documents through browser extensions with broad OAuth permissions.

They're not trying to cause a breach. They're just trying to finish their work faster.

But intent is irrelevant when sensitive data has already left your control. Shadow AI - the unsanctioned use of generative AI tools, browser extensions, and AI-connected applications - has evolved from an emerging concern into a concrete, board-level risk. And for third-party risk teams, the exposure doesn't stop at your own perimeter. It extends to every vendor, contractor, and service provider who handles your data.

What Shadow AI Actually Is (And Why It's So Hard to See)

Shadow AI isn't a single threat. It's a category of unmanaged third-party exposure that shows up in multiple forms:

• Standalone AI tools: ChatGPT, Claude, Gemini, and similar platforms used via personal accounts with no enterprise controls, retention policies, or data processing agreements
• AI-powered browser extensions: tools that summarize pages, assist with writing, or automate workflows, often with access to internal systems, ticketing platforms, and collaboration tools
• AI-connected applications: third-party apps authorized via OAuth to access Google Workspace, Microsoft 365, GitHub, Salesforce, or other enterprise systems
• Vendor-embedded AI features: AI capabilities quietly added to existing software-as-a-service tools your organization already approved, but without updated security review

The core problem is structural. Corporate AI accounts can be configured with enterprise controls: logging, retention limits, data residency, access restrictions, and contractual protections. Personal accounts cannot. When business data flows through an unmanaged tool, the organization loses visibility into where it went, how long it's retained, who can access it, and whether it was used for model training.

What was once manageable as an individual productivity quirk is now a systemic data governance failure.

Shadow AI Is a Third-Party Risk Problem

Most organizations have woken up to the internal dimension of Shadow AI. Employees need guidance, policies need updating, and monitoring needs to catch up with behavior. Fewer have internalized the harder truth:

Your vendors have the same problem, and it's your data at stake.

Every vendor that handles your customer records, source code, support tickets, financial documents, or credentials is a potential Shadow AI exposure point. If a vendor's engineer pastes your code into an unapproved AI tool, or a support agent summarizes customer complaints through a personal chatbot account, your organization inherits that risk, regardless of how mature your own internal AI policy is.

This reframes Shadow AI squarely as a third-party risk management (TPRM) issue. Unsanctioned AI tools are, in effect, unvetted sub-vendors, absorbing sensitive data through an employee's judgment call rather than a procurement process. They bypass the security reviews, data processing agreements, and contractual protections that your vendor program is designed to enforce.

The question your vendor risk program must now answer is no longer only: "Does this vendor protect our data?"

It's also: "Does this vendor know where our data is allowed to go next?"

The Incidents That Prove This Isn't Theoretical

Vercel, April 2026

Vercel disclosed unauthorized access to certain internal systems. According to Vercel’s announcement, the incident is linked to Context.ai, a third-party AI tool that was connected to an employee's Google Workspace account via broad OAuth permissions. This wasn't a case of someone carelessly pasting sensitive data into a chatbot. The AI-connected application itself became part of the enterprise access surface, a pathway into corporate systems created by a single employee's unsanctioned tool choice.

The lesson: AI-connected applications with broad permissions don't just absorb data. They can create attack vectors into the systems they're integrated with.

CB Financial Services, May 2026

CB Financial Services, parent of Community Bank, filed what legal commentators described as a landmark disclosure: the first SEC Form 8-K triggered by unauthorized employee use of an AI tool, not an external cyberattack. The company reported an internal incident involving non-public customer information processed through an unauthorized AI-based application.

Within weeks, multiple plaintiffs' firms announced investigations. Affected customers may pursue claims under negligence, breach of implied contract, invasion of privacy, and state consumer protection statutes.

The regulatory and legal signal is clear: unauthorized AI use is no longer just a policy violation. When regulated data is involved, it's a disclosure event, a governance failure, and a litigation risk.

Five Questions to Ask Your Vendors About Shadow AI

A Shadow AI vendor review doesn't require a 50-question questionnaire. It requires the right five questions, ones that quickly reveal whether a vendor understands where AI touches your data:

1. What AI tools do your employees use that may come into contact with our data, systems, or workflows?

This surfaces Shadow AI adoption inside the vendor's organization. Tools used for support, engineering, analytics, documentation, or customer success that your security team has never evaluated.

2. Do you have a formal policy governing employee use of generative AI tools, including explicit restrictions on customer data, source code, credentials, and confidential documents?

Vendor awareness and enforcement are different things. A vague "we follow best practices" answer is not sufficient.

3. Do your DLP controls detect sensitive data being pasted, uploaded, or submitted to AI tools?

Traditional data loss prevention was built before generative AI became ubiquitous. Many vendors' DLP coverage has meaningful gaps around prompt inputs and file uploads to AI services.

4. Do you monitor or restrict unauthorized AI browser extensions and AI-connected applications?

Browser extensions and OAuth-connected apps represent the invisible attack surface. Employees may not realize they're creating exposure by authorizing a productivity tool.

5. What data, including customer data, was used to train or fine-tune the AI models in your product, and under what contractual basis?

Vendors should be able to answer this clearly. If they can't, it's a signal that AI governance hasn't been operationalized.

The Bottom Line for Third-Party Risk Teams

Shadow AI isn't an edge case. It's a mainstream behavior happening inside your organization and every vendor organization in your ecosystem right now, at scale, largely invisible to your current controls.

The incidents from 2026 have removed any remaining ambiguity: this is not a theoretical risk. It's a data exposure vector, a regulatory trigger, and an active litigation category.

For third-party risk teams, the mandate is clear. Shadow AI governance needs to be embedded into vendor due diligence, contract language, and ongoing monitoring, not treated as a one-time awareness exercise.