Liability: Who Pays When It Fails

When an AI system causes harm, the law has to pick a defendant among: the developer (trained the model), the deployer (put it in a product), the user (relied on it), or nobody (it's just autonomous). No settled doctrine answers this — so liability flows through the doctrines that already exist, stretched to fit:

• Contract — who promised what; indemnities and terms do more allocation than any statute
• Tort — negligence and product liability supply the default rules when contract runs out
• Statute — the AI Act adds regulatory duties, but regulatory violation isn't automatically civil liability

The practical baseline courts keep reaching: the entity that chose to deploy the system in this context bears the duty of competent use.