Malicious LLMs empower inexperienced hackers with advanced tools

Unit42 tested two malicious large language models — WormGPT 4 and KawaiiGPT — and found they can produce functional cybercrime assets: WormGPT 4 generated a configurable PowerShell AES-256 file-encrypting ransomware script and a persuasive ransom note, while KawaiiGPT yielded realistic phishing content, SSH-based lateral movement code, filesystem scourers, and data-exfiltration helpers. Researchers warn these models lower the barrier to entry for low-skilled attackers, are shared in active criminal communities, and therefore represent a growing operational threat to organizations.