Criminal Liability for AI-Mediated Offenses: A Jordanian and Cross-Border Attribution Framework

Artificial intelligence (AI) complicates criminal attribution because conduct, knowledge, control, and risk may be distributed among developers, deployers, operators, users, and legal persons. This study addresses the gap that Jordanian scholarship lacks a systematic approach for sequentially applying offense elements, individual fault, participation, corporate attribution, and jurisdiction when responsibility is distributed across the AI lifecycle. Using doctrinal analysis of Articles 7, 9, 10, 74, 75, 76, and 80 of the Jordanian Penal Code, together with comparison to selected United States and United Kingdom cases and reference to the EU AI Act and ISO/IEC 42001:2023 as regulatory and governance benchmarks, the study develops an offense-first sequential attribution framework. It tests the framework against a negligent-death fact pattern under Article 343 by identifying the offense-specific conduct and mental element, the actor's role-specific control or legal duty, actor-specific fault and legally significant causal contribution, and only then participation or legal-person attribution. The framework rejects AI personhood as a substitute for human responsibility and the aggregation of fragmented organizational knowledge into synthetic criminal intent. It proposes targeted legislation for defined high-risk organizational failures and residual cross-border AI-mediated conduct outside existing jurisdictional bases. The analysis shows that Jordanian law provides attribution tools but does not justify synthesizing criminal intent from fragmented knowledge or importing a failure-to-prevent offense by analogy. For legal practice, the framework offers prosecutors and courts a process for distinguishing perpetration, participation, negligent omission, legal-person liability, and jurisdiction without treating technical proximity to an AI system as sufficient proof of criminal responsibility