Cybercrime and Dark AI

AI tools purpose-built or repurposed for cybercrime activities, including jailbroken models with no safety guardrails, are becoming an increasingly visible feature of online illicit markets. These products, colloquially referred to as Dark AI, are the collection of AI tools, services, and guides sold through online marketplaces and distributed in cybercrime communities. This talk will present initial findings from a digital ethnography of Dark AI as an emerging online illicit market in dark web, encrypted messaging application, and social media environments. Early themes include the use of AI to upgrade conventional illicit market offerings, the diversification of Dark AI products and services, and how these tools become normalise and framed in cybercrime communities. This research helps advance new approaches to AI safety by shifting attention away from downstream responses (e.g. victim education) to the upstream market conditions and socio-technical networks that enable the criminal exploitation of AI.