AI is showing up inside companies faster than anyone can track it. It is not just the chatbot an employee downloaded on their phone. It is also the AI features that show up automatically inside software the company already pays for, like a spreadsheet tool or a customer service platform that quietly turns on a new AI assistant during a routine update. Most companies have no formal way to catch either kind.
This gap already has a price tag. IBM's 2025 data breach report found that one in five organizations has suffered a security incident linked to unapproved AI use, and those incidents cost about 670,000 dollars more on average than a standard breach. Almost every company that got hit had no proper controls on what the AI could access. That is not a future risk. It is already showing up on the books.
The Samsung case from 2023 is still the clearest example of how this happens, and it is worth remembering because the mistake is so ordinary. Engineers pasted company source code into ChatGPT to get help fixing it, and the moment they did, that confidential code left the building. Nobody broke in. Nobody stole a laptop. An employee just tried to get their job done faster, and the company lost control of something it could never get back.
That is the core problem with AI inside a company. It is rarely a hacker breaking down the door. It is normal, well-meaning employees using a tool that seems harmless, connected to systems nobody checked. The tool doing damage is often authorized, just unsupervised.
Newer AI agents raise the stakes again. These are systems that can take actions on their own, not just answer questions: writing code, moving files, making changes without someone approving every step. When an employee runs one of these through a personal account instead of a company one, there is often no record of what it touched or why.
Boards of directors have started paying attention. Close to half of the largest US companies now name AI risk as something the board actively tracks, three times more than the year before. That shift usually means one thing: insurance companies, auditors, and regulators are starting to ask sharper questions, and "we did not know employees were using it" is no longer an acceptable answer in a hearing or a lawsuit.
The fix is not complicated, even if it takes real work. Companies need an actual, current list of every AI tool touching company data, not a survey from last year. They need someone with the authority to shut a tool off the moment it misbehaves. And they need to ask every software vendor exactly what AI features are getting added to their products and when, because right now most companies find out after the fact.
Waiting is the expensive option. The businesses that build this visibility now will pay far less than the ones that find out through a breach report.