AI computing power becomes new target for cyberattacks

Cybercriminals are increasingly targeting the computing power of generative artificial intelligence through a technique known as ‘LLMjacking’. According to research from Fortinet and FortiGuard Labs, attackers are shifting away from traditional methods like cryptocurrency mining or data theft in favor of exploiting stolen cloud credentials to access expensive AI models.

In an LLMjacking attack, hackers use compromised Identity and Access Management (IAM) keys or stolen login credentials to access legitimate cloud-based AI services. The goal is to force the victim's account to pay for the high-cost usage of high-performance models, which the attackers then use for themselves or resell to third parties. This method is highly attractive to criminals because it requires low investment in tools but yields high returns.

Research by Sysdig indicates the financial impact can be massive; for instance, the unauthorized use of models like Claude 3 Opus can result in damages exceeding $100,000 per day. Once a cloud identity is compromised, attackers can act within minutes to exploit the resources.