Source: LawGratis

LawGratis explores the complex and evolving legal theory surrounding criminal liability for autonomous systems. The blog post, dated December 14, 2025, questions whether current criminal law concepts of mens rea (guilty mind) and actus reus (guilty act) can be applied to non-human entities that operate independently of direct human control. It discusses hypothetical scenarios where an autonomous vehicle or a security drone causes harm or death, analyzing whether the liability should rest with the user, the manufacturer, or the software developer.

The author argues that the "black box" nature of deep learning creates a liability gap, as it becomes difficult to prove that a human developer "intended" or "foresaw" a specific harmful outcome generated by an evolving algorithm. The article suggests that legal systems may need to introduce a new category of "strict liability" for high-risk autonomous AI or create a "personhood" fiction for AI assets to attach insurance claims and criminal fines. It concludes that without legislative reform, prosecutors will struggle to hold anyone accountable for autonomous crimes.