Every AI decision turns into one of two things: a debt or a dividend. Which is this one?
In early 2023, Samsung leadership let its engineers use ChatGPT at work. Reasonable call. Their people were clearing bugs against deadlines that do not move, and here was a tool that could read a thousand lines of code and find the flaw in seconds. A real coup. Right?
An engineer, stuck on a bug that would not give, did the obvious thing. He pasted the confidential source code into ChatGPT and asked for help. It worked. It happened more than once. Nothing about it looked like a problem. The bug closed. The deadline held. Everyone went home.
Then it surfaced. No breach alarm; there was none to trip. Just the realization that the code now sat on outside servers Samsung could not reach or delete, and that nothing, no rule, no gate, no review, had ever stood in its way. By May, Samsung banned generative AI on company devices and set out to build its own.
Samsung is not a careless company. It is one of the most technically capable organizations on earth, staffed by engineers who understand exactly what a public system does with what you feed it. So if it happened there, this is not an intelligence problem. It is a leadership problem.
Don’t Blame the Engineer
Wave that off as one careless engineer, and you miss it. No manager believes their own engineer would put proprietary code into a public system. But that is what a good one does at 9 pm, against a deadline, reaching for the AI the company gave him to get the work done. His job was never really to “just” write code. It was to solve a problem with it, and he did.
The real failure sits above him. Leadership set the tool loose and set no rules behind it. Nobody said which data was off limits. There was no quick, legitimate way to ask the people who could say yes or no, security and the team whose data it touches, one question: am I allowed to use this here, and should it run on a private system, not the public one? No one was on the hook to answer. The easy path was to use it and stay quiet.
Urgency Is Not Speed
The story everyone tells is that companies are moving too fast. Slow down, freeze it. But urgency is not speed. Speed is how fast you move. Urgency is the pressure that talks you out of the step you should not skip. Samsung moving fast was fine. The debt arose from opening the tool before anyone agreed on how it would be reviewed, approved, and used. Who says yes? On what basis? Before it goes anywhere near your customers, your payroll, or your name. This is the conversation before you plunk down hard-earned money on a tool that might not be what you need.
Here is what pressure does. It does not make people reckless. It narrows what counts as the job. Under a deadline, a capable employee stops asking every question and starts asking one: what gets this finished. The other questions, who owns this, where does the data go, should someone approve it, do not feel skipped. They feel irrelevant. That is the part leaders miss when they go looking for the moment somebody chose to cut a corner. There was no such moment. There was a person doing the job as they understood it, inside a company that had not told them the job included asking. Speed is a measurement. Urgency is a lens, and it quietly edits what you can see.
The First Debt, and the One That Lets the Rest In
I call it the urgency debt. It is the first in the stack, and the one that lets all the others in. Every AI tool that goes live before there is an agreed way to review, approve, and use it is a decision made in the dark, real, owned by someone, but off the books and unseen by the people who should be on the hook.
Debt is borrowing from tomorrow to solve today. Financial debt borrows money. Technical debt borrows engineering effort. Urgency debt borrows governance, the decision about who approves this and on what basis, taken out against a deadline and not paid back. The bill arrives later, and it arrives with interest.
Not the Outlier. The Norm.
The numbers are not subtle. A 2026 Vanta analysis of more than 15,000 companies found that 70% have shadow AI running somewhere, tools employees brought in on their own. Only 2% ever went through a security review. And a separate 2026 EY poll of 500 technology leaders found that 52% of department-level AI projects were running without formal approval. Half.
Read the 70% slowly. Most of the AI making decisions inside your company walked in through a door that should have stayed locked until someone approved it.
Every organization believes it has time to write the rules later. That belief is where urgency debt is born.
What It Costs
The interest is specific. Someone did decide. There should always be an owner. But is it the right one? Without the conversation that decides who should own a call like this, ownership lands on whoever clicked, by default, not by design, made low and alone, a deadline breathing on it, nothing written down, no one else in the loop. When it goes wrong, and Samsung is proof it does, the company acts blindsided by a call one of its own people made. The tool stays, because pulling it means admitting no one was watching. The next person sees it running and assumes it was reviewed and approved. It was not. That is how one quiet decision becomes the way things are done.
Leadership signed off on the tool. No one signed off on how to use it.
The Bill You Cannot Refinance
And it compounds somewhere you cannot refinance: your people. Keep making them carry these calls alone, no cover, no rules, no one to ask, and they stop trusting the leadership that left them out there. The best ones leave. You meet that bill later, and it is the hardest to pay back.
But We Cannot Afford to Slow Down
Say that out loud in a leadership meeting and watch the room agree with you. Competitors are moving. Customers are asking. The person who calls for rules sounds like the person who wants to lose.
So look at what happened to Samsung. The company did not lose time to a review. It lost time to a ban. Every engineer who had been clearing bugs in seconds went back to clearing them the old way while the company built its own system from nothing. Whatever speed the tool bought, the cleanup took back with interest. That is not a company that moved too slowly. It is a company that moved fast in a direction it had not agreed on, then had to walk all the way back.
Be careful about the competitor you are picturing, too. The one running AI unwatched is not ahead of you. They may carry the same debt you do; they just haven’t been billed yet. You cannot see their interest accruing any more than they can see yours. Setting your pace by theirs is like judging your finances by a neighbor’s car.
Every Function Has a Stake in Who Says Yes
The fix is not a new committee or a freeze. Both are panic, not a plan. It is the leadership team coming together to decide who can review, approve, and use an AI tool, and who cannot, which matters as much. It must hold across every function, because they will all reach for these tools, and none can set the rules for the others. On what basis it can be used, where the data goes, and whether it runs on a private or public system. Who is on the hook when it reaches a customer or a paycheck? Give people a legitimate way to get that yes, so the next person with a tool and a deadline has somewhere to take it, instead of a call to make alone, and the answer is never “just use it quietly.” Then say it out loud to everyone, and keep saying it, because a rule nobody heard is no rule. A decision, written down now, or a cleanup later. If you don’t, you pay for the lack of one. The only question is whether you make it on purpose.
The Same Tool, as a Dividend
Now run Samsung’s story the other way. Same tool, same deadline, same engineer at 9 pm. Change one thing. Before anyone opened ChatGPT, leadership had already had the conversation: which data never leaves the building, whether it runs on a private system or the public one, and who to ask when you are not sure. Now, the engineer has a path instead of a guess. He just moves fast with a decision behind him, and the code never lands on a server Samsung cannot reach. That is the dividend: the same speed, without the leak, compounding for you instead of billing you later. The tool did not change. Someone decided on purpose.
And the dividend is larger than one leak avoided. In a company that has had the conversation, people know where to take a question, and they get an answer that week. Leaders know which seat owns which call. Security stops being the function that says no and becomes the one that says yes faster, because the criteria are written down and everyone has seen them. You approve the second tool quicker than the first, and the third quicker than the second, because you are not relitigating who decides every time. That compounds too. Not the absence of a mistake. The freedom to move with your whole weight behind you instead of one person’s guess at 9 pm.
Set your values before AI sets them for you. Urgency is what it looks like when you do not. Left alone, the tool chooses speed every time, because you never told it what matters more.
Where Am I Wrong?
Push back on me. Somebody reading this thinks the engineer should have known better, that this is a training problem and not a leadership one. If that is you, say so in the comments. That is the argument I most want to have, and the sharpest version of it ends up in the book.
Next week: the governance debt. You put AI to work, then everyone moved on. It did not.
Everybody’s using AI. The fix is to decide who gets to say yes before the tool goes live, not after the leak.
Want the whole picture, not just this one debt? Start with the AI Momentum Navigator; it shows you where your organization sits in readiness to use AI, scored against 8 dimensions.
New here? Subscribe to Debt or Dividend on LinkedIn for the next edition every Wednesday. Want more first? Everything I have written lives in one place: digitalBard Perspectives.