AI liability moves from law reviews to dockets

The question of who answers when an AI system causes harm is being answered the way common law answers everything: one case at a time. Early litigation over automated decisions, agent-initiated transactions, and model-assisted professional work is producing the first map of liability, ahead of statutes that remain largely aspirational.

The emerging pattern favors accountability along familiar lines. Courts have shown little patience for the-model-did-it defenses, treating deployed systems as tools for which deployers answer, while contract terms allocating AI risk between vendors and customers are receiving their first stress tests.

Corporate counsel are reading the early opinions as design guidance: documentation of testing, human oversight at consequential decisions, and honest marketing of capabilities all feature in the fact patterns that end well. The law, as usual, is arriving after the technology and before the legislation.