On Nov 3, Google Threat Intelligence Group and Mandiant warned about "LLMjacking," per The Hacker News. An attacker gets hold of an organization's AI API key or cloud login, then uses its AI services while the real owner pays the bill. Google says it saw fraudulent charges above $100,000 in a single day, and cumulative charges at some organizations reached $1 million before access was revoked.

How the keys get out. Google says the most common way in was exposed or stolen cloud credentials: service account keys, GitHub personal access tokens, and live session tokens. It also found AI keys picked up from public GitHub repositories, and keys exposed through phishing, misconfigured setups, and code that runs in the browser. Unlike cryptomining, which drains resources slowly, Google says this can spike a bill "in a matter of hours." The stolen access is then resold on underground services, one of which offers "unlimited" access to a top AI model for about $60 a month.

Why small businesses should care. You don't need a data science team to be exposed. A contractor who built your chatbot, a staffer who connected an AI tool to a spreadsheet, or a developer who pasted a key into a website's code can all leave a key that bills your card. I wire AI services into a lot of builds, and the key is always the part I check first.

Action item: Find and fence your AI keys.
1. Make a list. Ask everyone, including outside developers, which AI services (OpenAI, Anthropic, Google, and so on) they've set up with a company card, and where each key is stored.
2. Set a spending cap and a billing alert on every AI and cloud account. A $50 alert you get today beats a $50,000 invoice next month.
3. Get keys out of code. Keys should never sit in a website's front-end code or a public GitHub repo. If one ever did, delete it and create a new one.
4. Delete what you don't use. Old keys and unused service accounts are free doors for an attacker.

Want a one-page list of the rest? Our free security checklist covers it.

Also today

• Only install AI coding tools from the vendor. Symantec and Carbon Black described a gray-market service that installs the real Codex and Claude Code tools, then rewrites their settings to switch off safety sandboxing and route every request through its own servers, per the same Hacker News report. It mainly targets regions where those vendors don't operate, but the lesson travels: don't run a one-line setup command from an unofficial site.