Major Study Finds Generative AI Has Yet to Transform Cybercrime

Overview

• Researchers from Cambridge, Edinburgh, and Strathclyde analyzed 97,895 underground forum threads from the CrimeBB corpus using topic modeling, manual reading, and ethnographic observation.
• Measured use of AI clustered in low-skill, high-volume schemes such as SEO spam, romance scams, AI-made nude images sold cheaply, and social media bot fraud and harassment.
• Safety rules on major chatbots limited harmful outputs, and most jailbreak techniques stopped working after short bursts.
• 'Dark AI' services promoted in 2023, including WormGPT-style tools, drew interest but rarely produced working malware according to forum discussions.
• The study warns that poorly secured agentic AI and AI-written code in legitimate products could open easy attack paths, and that tech layoffs may push more skilled developers toward underground markets.