Google is warning that so-called LLMjacking attacks are leading to large fraudulent cloud AI charges after attackers gain access to credentials or exposed services tied to large language model infrastructure. According to the report, some organisations have faced losses of more than $100,000 per day, with total charges reaching around $1 million in certain cases.

What happened

The reported attacks involve criminals hijacking access to cloud-based AI resources, such as large language model endpoints or related credentials, and then using that access to generate high volumes of requests. The result is not only a security incident but also a direct financial one: businesses can be left paying for unauthorised AI usage.

Based on the source report, Google has described this as a growing abuse pattern affecting cloud AI environments. The issue is similar to other forms of cloud resource theft, where compromised accounts or poorly secured infrastructure are used to consume paid services at the victim’s expense. In this case, the target is AI capacity, which can be expensive when used at scale.

Why it matters for European businesses

For European companies adopting generative AI, the incident highlights a practical risk that goes beyond data exposure. AI services are often connected directly to pay-as-you-go cloud billing models, which means misuse can turn into a major cost event very quickly.

This matters especially for SMEs and growing digital businesses that are integrating LLMs into customer support, search, internal tools, content workflows or automation. If API keys, tokens or cloud environments are not properly secured, an attacker may be able to run unauthorised workloads without immediately disrupting the business. In some cases, the first visible sign may be an abnormal invoice.

The warning is also relevant for companies working with agencies, software vendors or internal development teams to roll out AI features. As more organisations embed LLM services into websites, apps and business workflows, the attack surface expands to include application code, CI/CD pipelines, cloud storage, developer environments and third-party integrations.

Who may be affected

• SMEs using cloud-based AI APIs for chatbots, assistants, automation or content generation.
• E-commerce businesses using AI for product content, support or search experiences.
• IT and development teams managing cloud accounts, model endpoints and API credentials.
• Digital agencies and software partners deploying AI features on behalf of clients.
• Founders and finance teams responsible for cloud cost control and vendor risk.

What companies should consider

• Review how AI credentials are stored and used. API keys and service account credentials should not be exposed in public repositories, client-side code or poorly secured environments.
• Enable billing alerts and usage monitoring. Fast detection is critical when AI services can generate high charges in a short period.
• Apply least-privilege access. Limit who and what can access model endpoints, cloud projects and billing-related services.
• Segment AI workloads. Separating environments and budgets can reduce the blast radius if one service is compromised.
• Audit third-party integrations and development practices. Businesses should understand whether agencies, contractors or vendors are managing AI credentials securely.
• Use anomaly detection where available. Unusual spikes in token usage, API calls or model activity should trigger investigation.
• Prepare an incident response process for cloud cost abuse. Security teams should treat suspicious AI consumption as both a cyber incident and a financial control issue.

For European businesses, the broader lesson is that AI adoption brings a new category of operational and financial security risk. Protecting data remains essential, but companies should also treat AI usage controls, cloud billing safeguards and credential management as part of core cybersecurity practice.