AI is becoming part of everyday work, often before businesses have had a chance to put formal policies around it. Employees may be using personal accounts or unapproved AI tools to summarize documents, analyze spreadsheets, draft communications, or work with customer and financial information. That creates a growing concern known as Shadow AI—when company data is shared with AI services the organization does not manage or fully understand.
The real issue is not whether employees use AI, but whether the business still controls its data when they do. Understanding the difference between consumer AI and a securely governed business AI environment—including how data is used, retained, protected, and whether it can be used for model training—is becoming an important part of responsible AI adoption.
AI is already showing up in the workplace. Employees are using tools like ChatGPT, Gemini, Claude and Copilot to write emails, summarize documents, research questions, analyze spreadsheets and save time on repetitive work. That can be a good thing. The problem starts when employees use AI tools the business doesn’t know about, doesn’t manage and hasn’t approved.
That’s called Shadow AI. And the real risk isn’t the AI itself. It’s what employees may be putting into it.
Picture a few everyday situations. Someone in accounting uploads a spreadsheet and asks AI to help spot a trend. An employee copies information from a customer document into ChatGPT and asks it to draft a response. A manager uploads an internal report and asks AI to summarize it. No one is trying to cause a security incident. They’re just trying to get work done faster. But in each case, company information has now been sent to an outside AI service.
That matters because employees are already bringing their own AI tools into the workplace. Microsoft and LinkedIn’s 2024 Work Trend Index found that 75% of knowledge workers surveyed were using AI at work, and 78% of those AI users said they were bringing their own AI tools to work.[1] CISA, the federal Cybersecurity and Infrastructure Security Agency, also warns users not to share sensitive or confidential information with AI systems, including workplace company data and personal information.[2] For dealerships, that could include customer information, employee records, financial data, contracts, internal business plans, or other information that was never meant to leave the company’s control.
One of the biggest misconceptions around AI is that choosing the tool is the same thing as securing it. It isn’t. A major difference exists between an employee opening a personal AI account and a business providing a company-managed AI environment. Take ChatGPT as one example. OpenAI says data sharing for model improvement is turned on by default for personal ChatGPT Free, Plus, and Pro accounts, although users can turn it off. For ChatGPT Business, Enterprise and OpenAI’s API platform, OpenAI says customer inputs and outputs are not used to train its models by default.[3] Same company. Similar AI experience. Very different rules around the data.
That’s why the better question isn’t: “Do we allow AI?” It’s: “Which version are we allowing, under whose account, and what happens to our data when employees use it?”
This is one of those phrases that gets thrown around a lot without much explanation. When someone says an AI company may “train on your data,” it doesn’t mean your uploaded file gets tossed into a folder where another user can simply ask for it later. The concern is more basic than that. An employee may submit information for one reason: “Help me answer this question.”
But if that service is also allowed to use the information to improve its future models, the data may now be used for another purpose. That may not matter much if the employee is asking about information that is already public. It matters much more if they upload customer records, payroll information, contracts, pricing, internal strategy, or other proprietary business information. The business should be the one deciding how that information can be used—not the individual employee clicking through an AI sign-up screen.
“Secure AI” does not mean there is a magical version of AI that can never be hacked, breached, or misused. It means the business has put controls around how AI is used and how company data is handled.
At a high level, a business should know:
Is our data used to train the AI? For sensitive business use, the answer should be clearly understood. Ideally, company data is not used to train a general-purpose model unless the business specifically agrees to it.
Who controls access? Employees should use company-managed accounts where appropriate, not random personal accounts the business can’t manage or shut down later.
How long is our information kept? “No training” does not necessarily mean “no retention.” The business should know whether prompts, files, and responses are stored and for how long.
Is our information separated and protected? Company data should be protected from other customers and unauthorized users.
What can the AI access? If AI is connected to email, documents, or other systems, it should only have access to the information it actually needs.
Can we manage and monitor it? The business should be able to set rules, control users, and investigate misuse if something goes wrong.
What has the provider agreed to do with our data? The contract matters. Businesses should understand what the provider can store, use, share or retain.
NIST takes a similar approach in its Generative AI Risk Management Profile, encouraging businesses to manage AI based on their own risks, business needs, and legal or regulatory requirements.[4]
Now imagine your dealership has done all that work. You’ve chosen an approved AI platform. Company data isn’t used for general model training. Access is controlled. Retention is defined. Employees know what they can and can’t do. Then someone opens a personal AI account and uploads a customer spreadsheet.
They haven’t just picked a different chatbot. They may have bypassed every protection the dealership put in place.
That’s the real Shadow AI problem. IBM’s 2025 Cost of a Data Breach research found that one in five of the 600 breached organizations it studied reported a breach linked to Shadow AI. Organizations reporting high levels of Shadow AI also experienced breach costs averaging $670,000 more than organizations with little or no Shadow AI.[5] Those numbers apply specifically to the breached organizations IBM studied, not businesses as a whole. But they show that Shadow AI is no longer just a theoretical concern.
AI can be useful. Employees are going to use it. The answer isn’t necessarily to block everything. The better approach is to give employees a safe way to use AI and make the rules easy to understand. Which tools are approved? What kind of information can employees put into them? What information should never go into an unapproved AI service? And who should employees ask when they aren’t sure? The most important question for business leaders is no longer:
“Should we allow AI?” It’s: “How do we let our people use AI without giving up control of our business and customer data?”
That’s the difference between simply using AI and using it responsibly.
LEARN MORE: For help with Secure AI for your dealership, please reach out to your PAA Member IntegraONE Account Manager, Chris Kurtz at ckurtz@integraone.com or call 484-223-3480 x2106.
REFERENCES
[1] Microsoft and LinkedIn. 2024 Work Trend Index Annual Report: AI at Work Is Here. Now Comes the Hard Part. May 8, 2024. Microsoft reports that the research included 31,000 people across 31 countries. (blogs.microsoft.com)
[2] Cybersecurity and Infrastructure Security Agency (CISA). Stay Safe Online When Using AI. CISA advises users to avoid sharing sensitive or confidential information with AI models, including workplace company data. (cisa.gov)
[3] OpenAI. Business Data Privacy, Security, and Compliance and current ChatGPT Data Controls documentation. OpenAI states that business-product inputs and outputs are not used for model training by default; for personal Free, Plus and Pro workspaces, data sharing is enabled by default but can be disabled by the user. (openai.com)
[4] National Institute of Standards and Technology (NIST). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1, July 26, 2024. (nist.gov)
[5] IBM and Ponemon Institute. Cost of a Data Breach Report 2025. The research was based on breaches experienced by 600 organizations globally between March 2024 and February 2025. (newsroom.ibm.com)