The $81,000 Side Project
One employee, one week, no fraud — and an $81,000 AI bill nobody was watching. The free-trial phase of AI is ending, and most companies have no controls built for what comes next.
HEADLINE STORY
The $81,000 side project
Last week a finance startup called Slash told its team to use AI tools more. One employee took that to heart. He spent about a day using an AI tool to build a goofy little video game. Every time he asked the AI to do more work on it, the cost ticked up, the way a taxi meter keeps running while you sit in traffic. Nobody was watching the meter. By Friday the tab read $81,267. The company is now rewriting its AI rules. To its credit, it is also promoting the game to make some of the money back.
It is an easy story to laugh at. Don’t. The number is the point. One person, one week, no fraud, nobody doing anything wrong. Just a tool that charges by how much you ask it to do, and no one watching the bill climb.
Here is the part that should land for any operator. The difference between companies on AI is no longer who is using it. It is who is controlling it. Ramp, a company that tracks corporate spending, found that the top one percent of companies now spend about $7,500 per employee every month on AI. The average company spends $11.38. Read those two numbers again. The leaders are not dabbling, and they are not letting it run loose either. And the same week that $81,000 bill went viral, one of the big AI providers ended its free trial and started charging everyone who had gotten comfortable using it for nothing.
The free-trial phase of AI is ending. Now everyone pays by the meter. Most mid-market companies have no controls built for that.
THE PATTERN
Three forces, one direction
Three things came into focus last week, and they all point the same way.
The free ride is ending. The trials are expiring and the AI providers are switching on the meter. AI is becoming a cost that can jump on you in a single weekend, not a flat monthly fee you set once and forget.
The best tools are getting harder to just buy. OpenAI released a powerful new tool for fighting cyberattacks last week and would not sell it to the public. You can only get it through the security company you already work with. More and more, the strongest AI shows up through the vendors you already pay, not through a website where anyone can sign up.
The power is pooling into fewer hands. Two of the top scientists at Google’s AI lab quit in one week and went to rivals. And the small group of companies that own the giant computer centers AI runs on keep signing the AI makers up as paying customers. Both the brains and the machines are concentrating with a handful of players.
The thread tying it together is simple. The stretch where you could just turn everyone loose and let them play is ending. What wins now is decision discipline: knowing what you are spending, on what, through whom, and why. The companies pulling ahead are not the ones using AI the most. They are the ones making the most deliberate calls about it.
WHAT IT MEANS FOR YOUR BUSINESS
Three moves for the next 30 days
One. Put a spending limit and a sign-off step on AI before you hand it to more people, not after. That $81,000 bill did not happen because someone was careless. It happened because the company told people to use AI but put no cap on the spending. Set a monthly dollar limit per person, turn on alerts that warn you when spending spikes, and give one named person the job of owning the AI bill the same way someone owns every other line on your P&L. This is basic discipline, and it costs nothing to put in place.
Two. Tell your team not to trust what an AI reads on its own. Security researchers showed a new trick last week: a hacker can hide fake instructions inside the routine reports your software generates, and the AI will follow them, handing over access without anyone noticing. Normal security tools did not catch it. Around the same time, a group of national security agencies warned that AI-powered attacks are now months away, not years. The rule for your team is simple. A person checks before an AI is allowed to act on its own, especially on anything it pulled in from outside the company.
Three. Do not bet your operation on one provider’s promise. Google publicly promised its next big AI release would land in June. With days to go, it has not shown up. Another provider took its newest tool from free to paid in two weeks. Set up the way you use AI so you can switch from one provider to another without tearing apart how you work. Decision velocity comes from keeping your options open, not from tying yourself to one name.
WORTH NOTING
Five more that matter
OpenAI released a powerful new tool for defending against cyberattacks last week and is delivering it through the big security companies, names like Cisco and CrowdStrike.
What it means: The strongest AI protection will reach most businesses through the security vendor they already use, so ask yours what they are adding and when.
A small number of companies now own most of the giant computer centers that all AI runs on, and they keep signing the AI makers up as customers.
What it means: Here is the threat to you. You don’t own that factory, and your AI provider doesn’t either; they rent it too. So when the handful of owners raise the rent or change the rules, the price and reliability of the tools you run on move with them, and nobody asks you. It is the strongest reason not to lock yourself into one provider.
Two of the most respected scientists at Google’s AI lab quit in the same week and joined competitors.
What it means: When you pick an AI partner for the long haul, follow the talent, not the brand name with the longest history.
Google promised its next big AI release would arrive in June, and with days left it still has not shipped.
What it means: Never build a plan around an AI feature that has been announced but not actually delivered.
Big companies are clamping down on AI spending. Uber, Coinbase, and Walmart have all put tighter limits on what employees can spend this quarter.
What it means: They are not backing away from AI. They are putting guardrails around it, and you can copy the move before you need it.
ONE QUESTION WORTH ASKING
If one of your people ran up an $81,000 AI bill this week, would you know this week? Or would you find out at the end of the month, when the statement shows up?
