Autonomous AI Agents: Who Is Criminally Liable? (2026)

Autonomous artificial intelligence agents — systems able to make decisions and carry out actions without direct human intervention — raise a major criminal-law question: if an agent causes harm, who is liable? As criminal lawyers, we set out the current framework.

The Problem

Unlike software that merely executes instructions, an autonomous agent plans, decides and acts to meet an objective. If, in doing so, it carries out operations that cause harm or fit a criminal type, the question arises as to whom the act is attributed.

AI Is Not a Subject of Criminal Law

Spanish criminal law rests on the principle of culpability: only those who act with intent or negligence are liable. An artificial intelligence is not a subject of criminal law: it cannot be sentenced or held criminally liable. Liability must necessarily be sought in the people behind the agent.

Who May Be Liable

• The individual who designed, configured, deployed or instructed the agent, if they acted with intent or negligence (for example, omitting due controls).
• The legal person, under the regime of corporate criminal liability of Art. 31 bis of the Criminal Code, where the agent operates for its benefit and there was a defect of organisation or control.
• In cases of an unforeseeable and unavoidable outcome, criminal liability might not be found, without prejudice to civil liability.

A Framework Under Construction

The question is open and will evolve with European AI regulation and emerging case law. What is certain today is the starting point: criminal liability falls on people, individuals or legal persons, not on the machine.