Monday morning, my AI bill kept me up at night
On Monday morning, I opened my finance dashboard and saw that the AI account I use for generating client copy had cost 800 yuan more than the previous week. My first reaction wasn't “the price went up,” but “Has someone been running jobs on my account?” I felt that panic before I even checked anything. I’ve also been stuck at this exact point. If you have accounts, API keys, or team spaces, don’t wait for the bill to explode before checking—I’ve learned that the hard way.
First, the context: how these platforms get targeted
Fireworks AI has confirmed a security incident in which some customer information and unfinished models were accessed. Reports describe the resale of stolen model access as “LLMjacking.” One of the risks is a massive bill spike after account abuse. My friend Lin Yan, an independent consultant in Hangzhou, received an unusual-login email from the platform on the afternoon of April 10. She immediately suspended a shared link, preventing further calls through her team space. The email is only a clue; the useful move is comparing login records, usage records, and charges together.
Today’s replication: check three places, then spend a little to set a cap
The cost is about 0 to 100 yuan, and the first round of checks takes 20 minutes. The barrier is low: I only need to log into the site dashboard, review the bill, and find the security settings. First, open the AI platform’s “login history” or “activity log,” then compare devices, locations, and IP addresses from the past seven days with my own usage patterns. Second, check “usage and billing” for a single-day spike. Third, go to the “API Keys” or “access keys” page, delete any keys I don’t recognize, and create a new set. Then I enable billing-alert emails, lower my spending cap, and review team members’ permissions. This tool isn’t necessary for everyone. If I haven’t uploaded client data to the platform yet, testing it can wait; organizing my own account list is the more urgent first step.
Match the checks to my stage—don’t do everything at once
If I’m just starting out and only use AI to write a few emails, I’d start with a five-minute check: sign out unfamiliar sessions, enable two-factor authentication, and confirm billing alerts. I wouldn’t rush to buy a security service. If I already have one or two clients and have started sharing an account or connecting it to a website, I’d immediately rotate the keys, review call logs, and set usage limits by client. If I’m scaling and my team has more than five people, I’d divide permissions by role, review members and keys every month, and keep a spending-anomaly response sheet. I can start with one reminder email today, without waiting for something to go wrong.