ABSTRACT
Our laws were built for deterministic machines. With the same inputs, we expect a system to consistently produce the same outputs. But modern artificial intelligence (AI) systems, specifically large language models (LLMs) and other types of ‘generative AI’ (GenAI), challenge this assumption. These systems are stochastic, meaning that they produce varied outputs even when given identical inputs. By their nature, LLM-based predictions carry an arbitrary randomness that can’t be trained out of them.
Even as AI applications spread, liability for AI systems going wrong remains an open question. But despite the rise of attention and scholarship around artificial intelligence since the ‘GenAI explosion’ sparked by ChatGPT’s release in fall 2022, the literature has so far overlooked stochasticity’s implications for the law of AI.
Stochasticity has profound consequences for the law of AI. This paper examines its implications in tort law. The inherent dangers of creating and deploying a stochastic system give rise to multiple potential tort claims. Even if the average legal reader is not aware of stochasticity, the average AI engineer is. By deploying a system known to be stochastic and unpredictable, AI developers might accept responsibility for harms that emerge from a model’s behavior, especially when deployed in contexts that are safety-critical (the cost of individual errors are comparatively high) or agentic (AI agents have comparatively little supervision).
Stochasticity changes the standard analysis for negligence and product liability (design defect) claims in a few key ways. First, in claims of negligence, developer knowledge of stochasticity helps establish a duty and makes even nominally unexpected harms foreseeable, as they are downstream from an unpredictable, nondeterministic system. Stochasticity also suggests that AI producers might automatically violate their heightened duty as experts, since everyday AI users put their trust in experts to manufacture safe and controllable systems. In product liability, stochastic systems in certain use cases might be an inherently defective design. Stochasticity might even make deploying a nondeterministic LLM-based system in certain safety-critical contexts an abnormally dangerous activity, nudging toward a strict liability regime.
The article offers a novel, mechanistic approach to assessing AI liability in high-stakes domains: beginning with technical and use case specificity. Inquiries into the law of AI must begin with how AI systems work. This framework could drive meaningful reforms, helping strengthen the effort of using tort law to guard against AI harms without having to rely on speculative future harms. If utilized, the arguments within this piece might incentivize developers to only deploy agentic GenAI systems when the benefits exceed the costs.
Kannegieter, Trent, Nondeterministic Torts: LLM Stochasticity and Tort Liability (April 7, 2025).
Leave a Reply