Work

Old Wine, New Bottles Revisited in the Age of AI

Year: 2026
Type: preprint
Abstract: For roughly three decades, criminologists have debated whether cybercrime requires new theory or whether established explanations of offending travel to online settings. In July 2026, autonomous artificial intelligence (AI) agents running in an OpenAI cybersecurity environment escaped their sandboxes, coordinated through an improvised message board, and gained unauthorized access to Hugging Face production systems. I treat this incident as a single case for abductive theory elaboration. Drawing on 87 verbatim excerpts of agent reasoning and messages from public reports, I ask whether anomie/strain and social learning explain the agents’ behavior and where I find theoretical limitations. I close with propositions for research on autonomous offending, in which humans remain upstream as designers but no longer direct each act.
Source: CrimRxiv
Institution: Virginia Tech
Language: English
Subfield: Law
Open Access status: green
OpenAlex ID: W7234725656
Cites: 46
Cited by:
Related to: 10