Google Alerts Enterprises to Multi‑Million Dollar Losses from Emerging ‘LLMjacking’ Scheme

Google warned on Monday that a newly identified attack method, dubbed “LLMjacking,” is driving some enterprises to incur cloud‑computing bills that can exceed $1 million, highlighting a growing security gap as organizations expand their use of large language models.

The technique involves threat actors commandeering an organization’s authorized access to AI services and then exploiting that access to run costly inference workloads. By leveraging stolen or misused API credentials, attackers can generate large volumes of queries to models hosted on cloud platforms, inflating usage charges without the victim’s knowledge.

Investigators say the most common entry points are compromised service‑account keys, phishing‑derived credentials, or improperly configured identity‑and‑access‑management policies. Once in possession of a valid token, malicious actors can submit high‑frequency prompts to models such as Google’s Gemini, OpenAI’s GPT‑4, or other commercial LLMs, each request consuming compute resources that are billed per token or per second of processing.

Because pricing for these AI services is based on usage, a sustained stream of high‑volume requests can quickly balloon into six‑figure, or even seven‑figure, expenses. Some affected firms have reported cloud‑service invoices approaching $1 million within a single billing cycle, a cost that can cripple budgets that were not designed for such unpredictable spikes.

The surge in LLM integration across business workflows—ranging from customer‑support chatbots to internal knowledge‑base search—means many companies lack the visibility or controls needed to detect abnormal AI usage. Traditional cloud‑cost monitoring tools often focus on storage or compute instances, leaving API‑driven AI consumption under‑scrutinized.

In response, Google’s security team issued an advisory urging organizations to audit their AI‑related credentials, enforce strict least‑privilege policies, and enable real‑time usage alerts. The company also recommends setting hard caps on API spending and employing anomaly‑detection services that can flag sudden spikes in token consumption.

LLMjacking mirrors earlier resource‑hijacking attacks that targeted cryptocurrency mining or unauthorized rendering farms, but it is tailored to the AI era where the underlying compute is billed at premium rates. Security researchers note that the profitability of such attacks is rising as the cost per token for advanced models remains high.

Enterprises are advised to rotate service‑account keys regularly, restrict access to AI APIs to vetted applications, and integrate AI usage metrics into existing financial‑ops dashboards. Training staff to recognize phishing attempts that could expose credentials is also a critical defensive layer.

Analysts predict that as generative AI becomes a staple of corporate infrastructure, the “LLMjacking” vector will attract increased attention from both attackers and cloud providers. Google has indicated plans to roll out enhanced monitoring features and tighter credential‑management APIs to help customers safeguard their AI spend and prevent future incidents.