Reasonableness Standard for AI Operational Security Liability

Determine a legally actionable reasonableness standard for AI operational security within United States negligence law by specifying expected practices and benchmarks for securing AI systems, including large language models, against adversarial attacks and other security failures so that courts can assess liability for harmful outcomes.

Background

The paper examines how AI safety and security perspectives intersect with governance and liability and highlights the complexity of assigning liability for harms caused by AI systems, particularly in adversarial settings where industry best practices are unsettled.

In discussing negligence law in the United States, the authors cite work noting that courts often rely on a reasonableness standard to evaluate precautions taken. However, because AI operational security lacks widely agreed-upon norms, what constitutes a reasonable standard remains unclear, complicating regulatory, judicial, and practitioner decision-making.