Shilun Zhou on Criminal Responsibility in Black-box AI Crime: Re-interpretation of Foreseeability and Inspirations for Canadian AI Regulations (IJLCJ)

"Criminal responsibility in black-box AI crime: Re-interpretation of foreseeability and inspirations for Canadian AI regulations"

Shilun Zhou (PhD Candidate)
International Journal of Law, Crime and Justice, December 2026, Vol. 87
Published online: September 2026

Abstract: Perpetrators using artificial intelligence (AI) to commit crime challenge the traditional criminal liability framework centred on foreseeability. This article argues that such a challenge is particularly acute in AI systems with black-box algorithms: perpetrators may exploit the lack of algorithmic transparency to claim they could not anticipate AI's output, thereby avoiding liability for criminal consequences, exceeding their mens rea. To circumvent the issue, this article adopts a novel knowledge-based approach to reinterpret the notion of foreseeability, and distinguishes between strict and relaxed interpretations of foreseeability, depending on the perpetrator's expected knowledge of criminal consequences. The strict interpretation is evident in the traditional doctrine, which requires knowledge only when perpetrators are expected to manipulate the criminal consequences. The relaxed interpretation suffices if the perpetrator knows the black-box AI will execute its programmed instructions without refusal, thus preventing them from evading responsibility by appealing to algorithmic opacity. The justification for the knowledge-based interpretation derives from preserving the autonomy of humans. This article further argues that AI systems should be designed to prevent and refuse participation in actus reus and should be subject to technical regulation. It also puts forward nuanced recommendations for Canada to enhance AI governance and anti-fraud regulation.