What is Shadow AI? It’s the use of AI tools within a business that have not been authorised or approved for commercial use by any company directors.
We’re not talking about high grade AI applications either, but AI tools in everyday use such as ChatGPT or Claude. With OpenAI having over one billion users, the odds are extremely high that people in your organisation are using it.
The problem is when AI tools are used for company purposes without any oversight. A member of HR uploads a stack of CVs to get a shortlist by lunchtime, a salesperson pastes a customer’s details and order history into a chatbot to build a proposal, or a customer service adviser drops in a complaint to get a polite reply drafted.
These are all understandable reasons, each member of staff is trying to do their jobs more efficiently and professionally. The trouble is that nobody has decided what may be entered, where it goes or how long the provider keeps it, so those decisions are being taken by whoever happens to be at the keyboard.
Samsung Learned The Hard Way
Three years ago, Samsung’s engineers pasted in source code and the contents of an internal meeting into ChatGPT. Whilst the staff had been given permission to use ChatGPT, the seriousness of the information leaked onto a third-party platform meant the company still had to ban the AI tools and build its own.
For Community Bank, the ramifications were more serious. In May of this year, the company discovered that customers’ personal information, including names, dates of birth and Social Security numbers, had been processed using an AI application the bank had never authorised. Within days, its parent company had determined the incident was serious enough to report to the US Securities and Exchange Commission.
Its banking systems had not been brought down by hackers, but an unauthorised AI tool had created a problem serious enough to involve regulators, investigators and senior management, which is precisely the danger of shadow AI.
Why It Becomes a Commercial Problem
Once staff start putting customer or employee information into unapproved AI tools, the business loses control over where that information goes and who has access to it. If a customer raises a complaint or an employee makes a data subject access request, finding out what happened becomes a time-consuming and potentially expensive exercise.
The business then has to establish what information was shared, which AI provider received it and whether it is still being held. That means diverting staff from their normal work, potentially paying for external advice and leaving directors to explain why nobody knew this was happening.
There is also the risk of regulatory investigation and enforcement action under data protection law. For companies operating or doing business in the EU, the EU AI Act adds further obligations, including transparency requirements for certain AI-generated content, with separate penalties for breaches.
Why Banning It Is the Wrong Answer
A blanket ban simply creates another problem, stripping employees of tools that save time and improve efficiency, while encouraging some to continue using them through personal accounts, or worse, their mobile phone.
I would rather businesses enable useful tech on terms they can defend than obstruct it and hope.
My advice? Start by finding out which tools people use, including personal accounts, meeting assistants and the AI features inside software you already own, and record which departments use them, for what and with what kinds of information.
Just as importantly, give staff an approved alternative for everything you take away, because people will not give up a tool that saves them an hour unless something else replaces it.
This might mean that you have to build your own AI and host it on your own server – a more expensive solution perhaps, but certainly achievable, and one many companies are implementing.
Bottom Line
As a matter of urgency, compile a list of the AI tools your staff use, what personal information has gone into each, where it has ended up and who approved it.
If the honest answer is that nobody knows, then your employees are already making decisions about your company and customer data that you won’t be able to explain to a customer, a regulator or even your own board.
Don’t take that risk.
Andy Chesterman, Managing Director, Privacy Helper












