When a work project is due but company technology is causing headaches, you can’t blame a manufacturing employee for looking for a workaround. Maybe it’s switching to a personal laptop or smartphone to complete a task or communicate, or downloading time-saving software from an app store.
The temptation to mix and match personal and work tech is understandable, but don’t do it. Introducing outside technology to a secure environment creates immediate risks to data and systems. So-called “Shadow IT” — the use of any personal or unapproved technology used in the workplace — represents a digital hole in the dike.
In the age of artificial intelligence, this familiar cybersecurity problem takes on a new dimension. Shadow AI is tempting. AI chatbots and code-writing programs promise to improve efficiency, but introducing them into an otherwise secure environment without proper controls can create serious problems.
“Nobody does this to hurt the company, but people are using tools they’re not supposed to use,” said Michael Tanji, director of cybersecurity for MxD, the National Center for Cybersecurity in Manufacturing as designated by the U.S. Department of War. “They’re exposing data that shouldn’t be shared to systems they have no control over. Basically they’re practicing computer insecurity at scale.”
While Shadow IT has been around for years, generative AI dramatically lowers the barrier to employees introducing unapproved technology into the workplace. Workers can now upload confidential documents to third-party AI tools for analysis, or use AI to build code and even entire applications without their organization’s IT or security team knowing.
The risk isn’t just theoretical. For example, Microsoft says it has identified thousands of instances of tech workers downloading AI assistant tools that turned out to be malicious impersonations. The victims thought they were downloading an AI browser extension that let them access ChatGPT while working in other applications.
Instead, they were downloading a malicious program that spied on users and scraped data. “This data provided insight into internal applications, workflows, and potentially sensitive information that users routinely shared with AI tools,” Microsoft said. The company reported 900,000 installations, with activity detected across more than 20,000 business organizations.
Manufacturers have already seen what can happen when employees turn to unapproved AI tools. In 2023, early in the chatbot era, Samsung Electronics determined that employees were uploading company information into ChatGPT to solve problems and manage documents. But once proprietary data is shared externally, it can no longer be protected. Samsung banned the use of generative AI for three years until it determined how to employ AI tools safely.
Four questions for Michael Tanji
We asked Michael Tanji about the risks of Shadow AI and how manufacturers can manage the temptation to use it.
What is the problem with Shadow AI?
Michael Tanji: It comes down to control. If you email a file to the wrong person, you can ask for it back or ask them to delete it. You can’t recall text pasted into an AI tool.
The account type makes a difference. OpenAI says personal accounts may train on your content unless you opt out, while business accounts don’t train on inputs by default. People using Shadow AI are, by definition, on personal accounts, so unless you know for sure to the contrary, you should assume that information you enter may be exposed beyond your control.
Are companies aware of the risks?
M.T.: Most firms know the risk but struggle to act. From an IT perspective, something like locking down USB ports is straightforward, but many companies lack a full-time IT or security person to do it. Policy helps only if people read it and someone enforces it.
The situation gets worse when you talk about AI. IBM research in 2025 found that only 37% of organizations had policies to manage AI or detect Shadow AI. Only 40% used access controls on AI models and data. When employees see a gap between the tools they need and the tools their company provides, they take matters into their own hands.
What should the proper protocols be?
M.T.: Determine whether you really need AI, and choose the one tool you’ll use for official business. You should have no problems finding a comparison chart that will help you decide.
Write a clear and concise policy. Who can use AI, what for, and how? What is absolutely, positively forbidden, and what are the consequences of noncompliance? Train your people.
There are plenty of free resources that will teach the basics. Too many people treat AI like search: AI is not search. A key reason why companies don’t see value in their AI deployments is because they didn’t teach people how to use it effectively.
Monitor use not only to ensure compliance with policy, but to help control costs.
Ensure a human reviews all AI output. If something goes sideways, it’s a human being that is going to be held accountable. “The machine made me do it” isn’t an excuse that is going to carry any weight.
