ABSTRACT
Attributing liability for torts caused by artificial intelligence (AI) constitutes one of the central and most challenging issues in AI governance. Neither the expansion of product liability nor proposals to confer legal subjectivity on AI systems provide a coherent framework for addressing AI-related torts. Legal risk must be distinguished from mere danger: the legal order cannot justify the imposition of strict liability solely on the basis that AI activities are deemed ‘high-risk’. As a general principle, AI-related torts should remain governed by fault-based liability, which allows for a nuanced and comprehensive assessment of the conduct of relevant parties while accommodating the diverse compensation claims of victims. The scope of fault presumptions should not be unduly extended. In determining liability, the integration of substantive rights-andduties regimes that enhance transparency with procedural rules on evidence disclosure can help to balance the litigation capacity of both sides and alleviate the evidentiary burden placed on victims. Even where an AI user’s conduct directly results in harm, providers may not automatically rely on safe-harbor provisions or contributory negligence defenses to claim mitigation or exemption from civil liability. To encourage the growth of an open-source AI ecosystem, the legal order should, as an exception, recognize liability exemptions for non-commercial, fully open-source AI systems-even though open-sourcing may increase the risks of misuse or malicious exploitation.
Wu, Hong, Fault-Based Liability for Artificial Intelligence Torts (January 21, 2026).
Leave a Reply