The Rise Of AI-Driven Malware: Threats, Myths, And Defenses

Overview: Generative AI models, such as ChatGPT and Claude, are increasingly leveraged by cybercriminals to create malware, but how severe is the risk for organizations? This talk explores the real impact of AI-based malware and explains why it doesn’t mean the end of cyber defense. We will delve into the distinctions between AI-powered, AI-generated, and AI-supported threats. We’ll illustrate these with examples of AI-generated threats that are still detectable. We’ll also cover polymorphic/metamorphic threats, such as ChattyCaty, which uses LLMs to dynamically alter its code. Following this, we’ll exa Details: mine what a fully autonomous Terminator threat could look like and discuss the EyeSpy PoC. Creating malware through AI represents an evolutionary, not revolutionary, step – so detection with existing methods remains viable. We’ll briefly demonstrate how we developed a Transformer model called EagleEye to effectively detect malicious behavior sequences. Finally, we’ll explore the emerging field of indirect prompt injections, which can enable novel spreading and data exfiltration threats, particularly through systems like RAG and CoPilot.