Acceleration Systems
Alerts & Business

Stolen AI Login Ran Up a $1 Million Bill Before Anyone Noticed

November 7, 2026 · Acceleration Systems

A stolen login used to mean a drained bank account or a few fraudulent purchases. Now it can mean a six-figure cloud bill run up in a single day — and the victim often doesn’t find out until the invoice arrives.

Google’s security researchers — the Google Threat Intelligence Group and Mandiant — published a report on November 2, 2026 describing a wave of attacks they call “LLMjacking.” The short version: criminals steal the credentials that let a business connect to AI tools, then use that stolen access to run enormous volumes of AI requests — on the victim’s dime.

What’s actually happening

Every business that uses cloud-based AI — a ChatGPT-style assistant wired into a help desk, an AI feature built into internal software, a developer calling an AI model’s API from code — has some kind of digital key that authorizes that connection. Google found attackers stealing those keys in several ways:

  • Leaked tokens in code. A developer accidentally commits an API key or access token into a public code repository, and attackers scan for exactly that.
  • Phishing and fake login pages. One group built a fake sign-up page for a legitimate AI service that tricked people into pasting in their active login session, bypassing passwords and even multi-factor authentication.
  • Exposed AI servers. Some businesses run their own AI software (like the open-source tool Ollama) on a server that’s accidentally reachable from the public internet, with no lock on the door at all.

Once attackers have a working credential, they don’t quietly siphon off a little value over months the way old-school “cryptojacking” (stealing computing power to mine cryptocurrency) did. They move fast: spinning up high volumes of AI requests, switching to expensive premium models, and in some cases creating and deleting backup accounts to keep access alive even after the first key is found and shut off.

Why this matters to a business like yours

This isn’t a story about giant tech companies only. Google’s report lists victims across financial services, retail, technology, and transportation companies in North America, South America, and Asia-Pacific — ordinary companies that simply had a credential leak somewhere.

The numbers are what make this different from typical fraud:

  • Individual incidents produced over $100,000 in fraudulent charges in a single day.
  • At some organizations, charges kept piling up until they reached $1 million before anyone caught it and cut off access.
  • In one case, an attacker made more than 500,000 AI requests in under eight hours using a stolen key, then resold that access to other criminals through an underground service — essentially running a black-market subscription business on someone else’s account.

That last detail matters: stolen AI access has become something criminals resell to each other, advertised on criminal messaging channels at a fraction of the normal market price. That’s a financial incentive to steal AI credentials that didn’t really exist two years ago, and it means a leaked key doesn’t just get used once — it can get sold on.

How victims find out — and how slow that usually is

According to Google’s report, these incidents are typically caught one way: someone finally looks at the cloud or AI bill and notices it’s wrong. The AI provider doesn’t usually warn you mid-attack. There’s no popup saying “we think someone else is using your account.” The charge just accumulates, invoice by invoice, until a finance person or IT admin asks “why did our AI spend jump 50x this month?”

By the time that question gets asked, the damage is already done. The fix isn’t detection after the fact — it’s catching the spike within hours, not at the end of a billing cycle.

What to actually do about it

You don’t need a security team to close most of this gap. A few concrete habits cover the attack paths Google describes:

  • Treat every AI and cloud API key like a password. Never paste one into code that gets committed to GitHub or any shared repository, even a private one that might later go public.
  • Turn on billing alerts. Every major cloud provider and most AI platforms let you set a spending threshold that emails or texts you when usage crosses it. This is the single most effective catch, because it’s often the only signal anyone gets.
  • Rotate keys when someone leaves. If an employee or contractor had access to an AI or cloud account, revoke or regenerate those credentials the day they leave — not whenever someone remembers to.
  • Don’t expose AI servers to the open internet. If your IT team or vendor runs an AI tool like Ollama on a server, make sure it requires a login and isn’t reachable by anyone who finds the address.
  • Check unfamiliar charges immediately. If a cloud or AI invoice looks off, don’t wait for the next billing cycle to investigate — hours matter here, not weeks.

None of this requires deep technical expertise — it’s the same instinct as freezing a credit card the moment a charge looks wrong, applied to a part of the business that’s newer and gets far less scrutiny.

For more on protecting the AI tools your business already relies on, see AIWAS’s knowledge base. Follow @AIWASai on X for the next alert.

Stop leaving customers on read.

We answer your reviews, keep your social pages posting, and make your ads convert, so you get more customers while you run the business.

Get Started