Black Box as a Justification for Strict Liability for AI-Related Damage

Strict liability is increasingly recognised as an appropriate framework for governing high-risk artificial intelligence (AI) systems, particularly those with ‘black-box’ characteristics, where internal operations are opaque and difficult to interpret. The inherent complexity of AI, including strong black-box features and unpredictability post-deployment, challenges the applicability of traditional tort law, which relies on establishing fault or negligence. Strict liability provides a means to hold entities accountable, addressing the difficulties in attributing fault in AI contexts. This work evaluates the merits and drawbacks of strict liability, explores its implications within the general liability regime, and provides concrete examples of AI-related harms that support this approach. The principle of AI neutrality and the persistence of fault-based elements within ostensibly strict liability frameworks like the Product Liability Directive are also examined, underscoring the complexities in regulating AI. Serbian legal doctrines regarding dangerous objects and activities provide courts with flexibility to adjudicate AI-related damages. Judges must comprehend the nuances of AI, including distinctions between traditional deterministic software and AI exhibiting emergent behaviour. While strict liability is beneficial for victim compensation and risk management, it can also stifle innovation and impose burdens on small enterprises. A balanced approach is essential to manage AI-related risks while promoting innovation.