The Dual-Use Dilemma of AI: Malicious LLMs

This Unit 42 report analyzes two malicious LLMs—WormGPT 4 (commercial/subscription) and KawaiiGPT (free/open-source)—showing they intentionally remove safety controls to produce high-quality phishing/BEC content, ready-to-run ransomware PowerShell scripts, data-exfiltration and lateral-movement code, and ransom notes; the models are distributed via Telegram, underground forums and GitHub, dramatically lowering the skill and cost barriers for cybercriminals and accelerating the scale and speed of attacks.