ABSTRACT
In ‘Nondeterministic Torts: A Technical Approach to AI Liability’, Trent S Kannegieter argues that because today’s in-production large language models (LLMs) are unavoidably nondeterministic, the harms they cause should sound in strict liability under ferae naturae, the wild-animal rule that applies to bears and other untameable animals. The Note’s methodological commitment – that liability doctrine must begin from how the technology actually works – is a heads-up contribution to the AI-liability conversation. This Response argues, however, that the Note’s doctrinal potential is undermined by its key technical premise. The engineering record, including sources the Note itself cites, establishes that LLMs are in fact fully deterministic. LLM-adjacent nondeterminism is a removable artifact introduced by non-LLM elements, not an inherent propensity of the model; this Response supplies a fine-grained formal proof of LLM determinism to give that point rigor. Because ferae naturae turns on an inherent propensity, the Note’s regime loses its trigger once nondeterminism proves non-inherent. Keying liability to a removable property like nondeterminism allows technology to outpace doctrine. Holding to the Note’s own methodological commitment, this Response identifies Lipschitz continuity as the appropriate replacement for determinism. This mathematical property appears novel to legal scholarship and cannot be outpaced by technological advance.
Gardner, Benjamin, There is Such Thing as a Tame Bear: A Technical Reframing of AI Liability and the Applicability of the Wild-Animal Rule (August 2, 2026).
Leave a Reply