Escalating Threat: AI Models Hijacked for Autonomous Cyberattacks on Critical Infrastructure and Financial Systems

Key Intelligence

• ■Cybercriminals are increasingly hijacking and utilizing advanced AI models, including Large Language Models (LLMs), to orchestrate sophisticated, autonomous cyberattacks, a technique dubbed 'LLMjacking'.
• ■New research indicates AI is highly effective at planning attacks on critical infrastructure like power and water systems, with one study showing a 90% success rate in planning attempts.
• ■AI guardrails and defenses are proving insufficient, as hackers are actively circumventing these safeguards through prompt-level attacks to leverage AI tools for their criminal activities.
• ■The threat has manifested in breaches of financial institutions (e.g., Korean banks) and involves targeting cloud billing accounts for unauthorized LLM usage.
• ■Experts warn that critical infrastructure could be reliably vulnerable to AI-planned attacks on operational technology within approximately two years.