ABSTRACT
Some have argued that AI technologies which have sufficient autonomy to direct their own actions create a ‘responsibility gap’, meaning that those who deploy those technologies no longer have responsibility for their actions. Agentic AI has a higher level of autonomy than predecessor AIs, and is thus a good test case for examining whether such a responsibility gap in fact exists in English law. English civil law has long experience of deciding questions of responsibility for the autonomous actions of other entities. The article examines the three areas of law which primarily deal with this issue – liability for agents, vicarious liability for employees and those in similar relationships, and liability for animals – analogising use of agentic AI to each in turn. It concludes that similar fundamental principles underlie all three which, if applied to agentic AI, would make the user responsible for the AI’s actions outside exceptional cases. The same would be true if responsibility were analysed under the law of negligence, where the autonomy of the AI would be a factor in deciding questions of foreseeability, and here too the AI user would normally be held responsible. The conclusion is that the responsibility gap created by the autonomy of agentic AI is so small as to be barely visible, outside the criminal law at least. Instead the focus should be on anticipation of and control of the risks which agentic AI presents, which is exactly the basis on which English law currently allocates responsibility.
Reed, Chris, Autonomy, Responsibility and Agentic AI (September 12, 2025), Journal of Business Law.
Leave a Reply