Stolen AI access, sudden cloud bills
Google’s Threat Intelligence Group (GTIG) is warning that a form of cybercrime known as LLMjacking is leaving cloud AI customers with enormous bills. In these attacks, criminals take over company AI accounts or cloud credentials and run expensive artificial-intelligence workloads that the victim ends up paying for. GTIG says the charges can climb past $100,000 within minutes.
Unlike cryptojacking—where stolen computing power is used to mine digital coins and costs often rise more slowly—LLMjacking can inflict financial damage almost immediately. For Pakistani firms, freelancers, and startups that already rely on pay-as-you-go AI APIs (application programming interfaces that let software talk to AI models), that speed matters: a leak can turn into a billing shock before anyone notices.
How attackers get in and cash out
According to Google, hackers usually start by hunting for exposed AI endpoints or by stealing credentials. Those credentials can include API keys, identity tokens, service-account keys, GitHub personal access tokens, and active session tokens. Once inside, they try to raise their privileges, keep a foothold, and locate valuable AI resources.
The commercial aim is clear. Attackers run their own AI jobs on the compromised account, push requests through the hacked access, or sell that access to others. Google describes the pattern as monetisation and resource-hijacking of foundation models—large general-purpose AI systems—through methods such as token proxying, automated model streaming, and GPU (graphics processing unit) provisioning for heavy compute.
In one case cited by Google, a group labelled UNC6993 obtained LLM (large language model) API keys, switched on services in a victim’s cloud project, and generated more than $100,000 in charges. In another pattern, attackers enrolled compromised accounts on pay-as-you-go plans, selected premium AI models, and launched large volumes of inference work—the step where a trained model answers queries. To dodge rate limits, they repeatedly created and deleted temporary service accounts while watching the victim’s billing dashboard.
A growing specialty in cybercrime
Google says LLMjacking is becoming a specialised trade. Some crews focus on breaking into company systems. Others, called “LLM Mixers,” operate gateways that gather stolen credentials, forward requests, and sell unauthorised AI access to buyers on underground markets.
GTIG expects the problem to intensify as the market for AI services expands. Google’s guidance is practical: organisations should monitor their AI platforms closely, treat any unauthorised change to AI accounts or services as a serious security issue, and apply the principle of least privilege—giving people and machine accounts only the permissions they truly need.
For teams in Pakistan that store keys in code repositories, share cloud consoles widely, or leave AI endpoints open to the internet, the same habits that once caused smaller cloud surprises can now produce six-figure losses in a very short window.
Aasan alfaaz mein
Google ke mutabiq hackers ab companies ke AI accounts aur cloud passwords chura kar mehengi AI services chalwa rahe hain. Is LLMjacking se bill kuch minutes mein $100,000 se zyada ho sakta hai, jo company ko dena parta hai. API keys aur cloud access ko tightly control karna zaroori hai. Agar AI account mein bina ijazat change dikhe to usay foran security masla samjhein.
