Modern technology has evolved beyond what was previously imaginable. Artificial Intelligence is the next big step in our technological evolution as AI has made its way into phones, robots, and now our vehicles. 94% of vehicle crashes today are due to human error. But what happens when you take the human out of the equation, and we rely on these automated systems to navigate our streets?[1] Existing legal principles emphasize fault based on the person operating the vehicle, but who is to blame when companies begin promoting no driver necessary transportation?[2] This modern problem presents a legal dilemma as AI constantly evolves and currently lacks the intent to be liable, while also missing the required foreseeability in traditional tort cases. This article argues for a hybrid tort framework that blends strict product liability, an industry-calibrated ‘reasonable computer driver’ standard, and the traditional reasonable person benchmark. This approach ensures accountability while promoting innovation and public safety.
Background: Why Tort Law Doesn’t Fit
Traditional tort law relies on a checklist used to determine negligence. Specifically, a HUMAN can be deemed negligent based on duty, breach, causation, foreseeability, and harm.[3] These systems have fared well when applied to people but tend to blur when considering the negligence of automated systems. AI is complicated: though it is ultimately a product, it could think and decide similarly to a human being.[4] Under this framework, issues arise when considering whether a product liability analysis or treating AI as a legal person would more adequately provide liability. Specifically, AI is constantly evolving, much like a black box in an airplane.[5] Thus, product liability cannot be provided because although the product is absent of any defects when sold to the consumer, AI is evolving and prone to making mistakes.[6] Similarly, following the negligence checklist, foreseeability cannot be provided when programmers cannot predict the next actions of AI.[7] Because existing frameworks cannot allocate fault, scholars and policymakers have suggested several reforms to adapt to the presence of automated cars.
Strict Product Liability for Manufacturers
Since operation of some vehicles has shifted from human-operated to AI, courts may begin to look at strict liability to begin assigning liability. Traditional tort law places liability on manufacturers for manufacturing defects, design defects, or failure to warn, even when the manufacturer exercised all possible care.[8] In the context of automated cars, manufacturing defects occur when flawed prototypes are shipped by the manufacturers and becomes unsafe due to an error or flaw in how it was made.[9] Design defects occur when the manufacturer could not anticipate certain conditions that a reasonable person would expect when on the road.[10] Examples include emergency systems for pedestrians jumping out on the road or adjustable driving styles based on weather conditions.[11] Using a strict liability approach removes the need for negligence and places the burden on the expectation of the product. This ultimately means that the product — the AI operated cars— should meet the product standards expected by the public and is not inherently dangerous to use.
Although this standard is quite protective of society, it also has its drawbacks. A policy this strong in favor of public safety may deter innovation, as companies may not want to risk liability for large scale settlements where juries tend to vote unfavorably for these corporations. Furthermore, as AI continues to evolve, placing liability on manufacturers on an outcome it did not initially deem possible would yield unfair results. This has led scholars to adopt a new model, known as the reasonable human standard, along with the reasonable computer driver standard, which evaluates an autonomous vehicle’s behavior against what a competent AI system should have done under the circumstances.[12]
The “Reasonable Computer Driver” Standard
As scholars began to develop new ways to motivate innovation while protecting the public, the reasonable computer driver standard emerged. Here a jury would ask whether an AI system performed as safely as a competent autonomous driver would under the same conditions.[13] Rather than imposing a strict liability regime, we begin to treat the AI operator as a person rather than a product.[14] In doing so, the courts would compare the AI operator to other AI operators in the field and point out deficiencies that resulted in the accident that took place.[15] The standard would be set by industry norms and what is deemed to be competent at the time.[16]
Although this initially appears to be a feasible means to assign liability, companies can escape liability through the lack of advancement in the field. Today, there are not many self-driving cars in production, nor are there many companies in development of this product.[17] The limited information and inability to set a robust standard means that the defining threshold for what a reasonable automated driver could be is shaky at best and may allow manufacturers to escape any liability. Difficulty further increases when courts evaluate what a reasonable AI operator means, as great technical experience is required, especially when the algorithms’ decision-making processes are unclear even to their creators.[18] The absence of any precedent, and consensus on what qualifies as reasonable applies additional constraints when presented to the court. AI is constantly evolving even after sale by the manufacturer, so in that sense, determining what is “reasonable” begins to become blurry.
Reasonable Person Standard
This leads to the latest development of a hybrid model that imposes both a negligent product liability regime and evaluates the AI operator under a reasonable person standard.[19] The negligent product liability standard would apply to the product itself, and be specific to the product design and manufacturing defects.[20] This primarily holds manufacturers liable for the outcome of their products and ensures that the manufacturers are able to prove that their products were up to standards at the time of release.[21] It also allows for the baseline expectation that the product is viable and dependable prior to applying the human driver standard.
Courts are familiar with the negligence standard for a person, so this allows for easier application of the human driver standard to future cases. The AI operator is expected to make driving safer by removing the person out of the equation, so the bare minimum expectation is that the AI-driver can operate the vehicle as well as a person can. Liability attaches when an autonomous vehicle performs worse than an attentive, unimpaired human driver would have in the same situation. Inexcusable mistakes such as speeding in the rain or driving in the shoulder would not be tolerated. This framework keeps the focus on outcomes rather than technology, ensuring that AI systems meet or exceed human safety expectations. Ultimately, the human standard provides a stable baseline for courts which ensures accountability without letting technological limitations dilute safety expectations.
Conclusion
As innovation continues to develop, there will be nuances to existing frameworks that guide assigning liability. The law needs to strike a proper balance between accountability and innovation working together. Neither the strict liability standard, the reasonable computer driver, nor the reasonable person standard alone can dictate the best means for protecting the public. By integrating elements of negligent product liability and utilizing the reasonable human driver standard, we can promote safety, clarify liability, and ensure that as machines take on more responsibility on the road so human expectations and protections remain at the forefront.
[3] David G. Owen, The Five Elements of Negligence, 35 Hofstra L. Rev. 1671, 1674 (2007).
[4] Lea Pődör & István Lakatos, According to Whose Morals? The Decision-Making Algorithms of Self-Driving Cars and the Limits of the Law, 6 Future Transp. 5 (2026), https://doi.org/10.3390/futuretransp6010005.
Comments