A story is making the rounds again: an employee allegedly cost their company 70,000 euros through uncontrolled AI usage. Sounds like a horror story. And it's true — but only under very specific circumstances that you as a private user are unlikely to encounter.
The pattern feels familiar — just like with data privacy: yes, it happened. No, not the way the headline implies.
Quick reassurance upfront: as long as you understand how billing works, this won't happen to you.
How Do You Burn Through 70,000 Euros in Tokens?
Only if several things go wrong simultaneously.
You're using an API (not a subscription, but usage-based billing), you haven't set a spending limit — and then you let AI agents run largely uncontrolled. Agents (automated AI programs that complete tasks independently) can call each other, hold long conversations, send thousands of requests. And each one costs money.
That's not a bug. That's the design — for developers building large automated workflows. For private users using ChatGPT to write emails: not relevant.
No Payment Method Stored? Zero Risk.
Then literally nothing can happen. No credit card means no API, no rechargeable credit, no surprise bill.
And no, you don't need a credit card to try ChatGPT, Claude, Gemini, or Perplexity — all have free tiers that work without any payment details. We've already covered what's actually available for free if you want the full picture. Just sign up and start typing.
You're Using a Subscription?
Around 20 euros per month, works like a mobile data plan: when it's used up, it's used up — you wait until next month. "Subscription" doesn't mean "pay until we've had enough." It means: pause.
Some providers offer an optional buffer. Anthropic, for example, lets you set a limit for how much you can spend after your plan runs out. You get notified when you hit it. And you can set that buffer to zero — then nothing runs. You know this: mobile data auto-extension that you can switch off.
Worth noting if you do use the buffer: it costs significantly more per token than the subscription plan — intentionally. It's meant for when a project needs to finish and you can't wait. For private users: usually not necessary.
You're Using an API?
Here lies the theoretical risk. API usage is pay-per-use — you pay per token (a kind of syllable unit, explained here). But every reputable provider lets you set limits. And you should.
Rule of thumb: 5 euros per day or 100 euros per month. When the limit is reached, nothing more happens. You wait until tomorrow or next month.
A tip on limit strategy: prefer small limits over short periods rather than large limits over long periods. A monthly limit of 100 euros exhausted on the 3rd leaves you waiting 27 days. A daily limit of 5 euros leaves you waiting until tomorrow — much more manageable.
One exception: Google. Their AI product portfolio is somewhat harder to navigate and targets larger teams. For beginners: start with more straightforward providers.
From Personal Experience
I've used OpenAI, Anthropic, Mistral, MiniMax, OpenRouter, Cortecs, Perplexity, and GitHub Copilot myself — some still active, others not anymore. All of them respect the set limit or the subscription. Not a single surprise bill. Not even an unpleasant one.
The only scenario where things can explode: you've explicitly configured "use whatever you want" and then let agents run free. But that's a decision — not a nasty surprise.
Bottom Line: No Panic
Use the big, reputable providers. Only store a payment method when you actually need one. Get a subscription if you use AI regularly — 20 euros is fair for what you get. And if you're working with an API: set limits. Now. Before you start.
The 70,000-euro story isn't a failure of the providers. It's a configuration error without a safety net. And the safety net is exactly two clicks away.