The Prompt is the Payload – Weaponizing LLMs on the Dark Web

It started with curiosity. Could ChatGPT write phishing emails? Could it simulate customer service? Could it imitate corporate jargon?

The answers came quickly: yes, yes, and yes—and the dark web noticed.

In a matter of months, cybercriminal forums began experimenting with open-source large language models (LLMs) like GPT-J and LLaMA. Jailbroken clones of…