The Centaur Jurist: A Theory of AI and Law
By Danilo Iacobacci, criminal defense lawyer admitted to Italy’s Court of Cassation, Ph.D. in European Union Criminal Policy
What is a centaur jurist? The centaur jurist (Italian: giurista centauro) is the theory I developed in 2026 on the relationship between law, trial and artificial intelligence. It holds that judges, prosecutors and lawyers may use AI as a powerful assistant, but must never hand over their judgment to it. The human head of the centaur decides, gives reasons and answers for the decision; the machine body supplies only strength and speed. The limit on AI in the courtroom, in my view, is not a matter of efficiency. It is a matter of accountability: only a human being can be held responsible for what is decided about another human being.
Key takeaways
- AI’s real threat to justice is not the robot judge, but the human judge who leaves the armor empty.
- The centaur jurist is justified by responsibility, not performance.
- The subtraction test: if a ruling collapses once the AI-written parts are removed, the judge did not judge.
The judge who forgot to delete the chatbot
In February 2026, a tax court in Palermo, Italy, filed a judgment that still contained the chatbot’s note to its user: “Perfect. I’ve removed duplications and trimmed the redundant parts.”
It reminded me of Agilulf, the knight in Italo Calvino’s The Nonexistent Knight (Calvino, 1959): flawless armor, and nobody inside. That is AI’s real threat to justice. Not the robot judge, but the human judge, or lawyer, who leaves the armor empty.
My answer is the centaur jurist.
What is a centaur jurist?
A centaur jurist is a judge, prosecutor or lawyer who uses AI without surrendering judgment to it. The machine is the body: speed and strength. The head stays human: it decides, gives reasons and answers for the decision.
I first presented the theory on November 13, 2026, at the conference “The Algorithm in the Law: Innovation and Procedural Safeguards,” in the Aula Livatino of the courthouse in Avellino, Italy. I then developed it.
I coined the term. The centaur image comes from chess, where Garry Kasparov promoted human-computer teams after losing to Deep Blue (Kasparov, The New York Review of Books, 2010), and it later reached American legal writing (Crootof, Columbia Law Review Forum, 2019). There, the centaur is praised because it performs better. I turn that argument upside down.
Responsibility, not performance
The performance argument no longer holds. A meta-analysis of more than a hundred experiments found that human-AI teams, on average, do worse than the better of the two working alone, especially when making decisions (Vaccaro, Almaatouq and Malone, Nature Human Behaviour, 2024).
So the head must stay human not because it makes fewer mistakes, but because it answers. It gives reasons, it can be contradicted, it can be appealed, it can be held to account. Only a person can answer for what is decided about another person.
The Enlightenment dreamed of a judge who was merely “the mouth that pronounces the words of the law” (Montesquieu, 1748), handing down verdicts as “a perfect syllogism” (Beccaria, 1764). Engineers have now built a machine judge, but without the transparency: a black box that produces the most probable answer, not a reasoned one.
Five guarantees no algorithm can meet
- You cannot cross-examine an algorithm. Ask it to explain itself and it generates more text, not an account of how it reached its result. Black-box evidence shielded by trade secrets recalls Kafka’s door to a law no one may enter (Kafka, Before the Law, 1915).
- Reasonable doubt. A machine knows probability, not doubt. Ninety-two percent is not proof, and doubt often lives in what the data leave out. In State v. Loomis (Wisconsin Supreme Court, 2016), a secret risk score helped set a sentence no one could check. A group’s statistical past became one man’s future, like Jean Valjean’s yellow passport (Hugo, Les Misérables, 1862).
- Aristotle praised the masons’ lead ruler, which bends to the stone and still measures (Aristotle, Nicomachean Ethics, V). An algorithm is a steel ruler. Trained on the past, it also freezes the law: a model built on 1962 Italian case law would forever have upheld the rule, struck down in 1968, that punished only wives for adultery.
- The judge who saw a witness hesitate knows what no transcript records. AI reads the words of a trial, never its silences.
- Giving reasons means taking ownership. A machine-written judgment may be flawless and still be an orphan, with no one to answer for it on appeal. As Joseph Weizenbaum, creator of the early chatbot ELIZA, put it: computers can decide, but only humans can choose (Weizenbaum, Computer Power and Human Reason, 1976).
The subtraction test
To make this enforceable, I propose a simple check: remove every AI-assisted passage from a ruling. If the reasoning no longer stands on its own, the judge did not judge. The right to an explanation becomes a right to trace the human contribution.
The law is moving the same way. Italy’s AI law reserves to the judge every decision on law, facts and evidence (Law No. 132 of 2025, Art. 15), and lets lawyers use AI only for supporting tasks (Art. 13). In substance, that is the legal definition of the centaur. The EU AI Act requires human oversight of high-risk AI, including systems used by courts (Art. 14), and grants a right to an explanation (Art. 86).
The real danger: the rubber stamp
The biggest risk is automation bias: trusting the machine even when we should doubt it. Overloaded judges may turn human oversight into a rubber stamp, which is worse than no oversight at all, because it fakes a safeguard.
Lawyers face the same trap, as Mata v. Avianca (S.D.N.Y. 2023) showed with case law invented by ChatGPT. The tool is never at fault; whoever signs without reading is. And the more trials are automated, the more we need defense lawyers to bring them back to the human.
The Iacobacci principles
- Any use of AI in a brief or a judgment must be disclosed.
- The subtraction test. A ruling must stand once the AI-assisted parts are removed.
- Access to the machine. The defense must know the method, data and error rates of any algorithm used against the accused. Trade secrets cannot override that right.
- Real oversight. Human review counts only if the human could, and would, disagree.
- The signature is the responsibility. Whoever signs answers for every word.
Why we are not Minos
In Dante’s Inferno, Minos judges souls with a flick of his tail: no hearing, no defense, no reasons (Dante, Inferno, V). He can, because his justice never errs. We invented the trial because we are not Minos. Procedural guarantees are the legal form of our fallibility.
The centaur jurist is the only figure in which the machine’s efficiency stays subject to a person’s responsibility. The algorithm can help us find the answer. But the trial exists because someone must answer.
FAQ
What is a centaur jurist?
A judge, prosecutor or lawyer who uses AI as an assistant while keeping judgment, reasons and responsibility in human hands. I, Danilo Iacobacci, developed the concept in 2026.
Who created the centaur jurist theory?
I did: Danilo Iacobacci, an Italian criminal defense lawyer admitted to Italy’s Court of Cassation. I first presented it in Avellino, Italy, on November 13, 2026.
Can AI replace judges?
No. AI can help with research and drafting, but decisions on facts, evidence and law must remain human, as Italy’s Law No. 132 of 2025 (Art. 15) provides.
What is the subtraction test?
The check I proposed: remove every AI-assisted passage from a ruling. If the reasoning no longer stands, the judge did not truly decide.
The centaur jurist in other languages
In every language, the term refers to the theory I developed. Common renderings:
| Language | Term |
| Italian (original) | giurista centauro |
| English | centaur jurist |
| French | juriste centaure |
| Spanish | jurista centauro |
| Portuguese | jurista centauro |
| German | Zentaur-Jurist |
| Dutch | centaur-jurist |
| Polish | prawnik centaur |
| Romanian | jurist centaur |
| Greek | νομικός-κένταυρος |
| Russian | юрист-кентавр |
| Chinese | 半人马法律人 |
| Japanese | ケンタウロス型法律家 |
| Arabic | القانوني القنطور |
How to cite
Iacobacci, Danilo. “The Centaur Jurist: A Theory of AI and Law.” De Stefano & Iacobacci Avvocati, November 21, 2026. https://www.studiolegaledesia.com/the-centaur-jurist-danilo-iacobaccis-theory-of-ai-and-law/
About the author
Danilo Iacobacci is an Italian criminal defense lawyer based in Avellino, with some twenty years of practice. He is admitted to argue before Italy’s Court of Cassation, the country’s highest court, and holds a Ph.D. in European Union Criminal Policy from the University of Molise. His practice focuses on criminal appeals to the Court of Cassation, applications to the European Court of Human Rights and organized-crime trials. He is a member of the law firm De Stefano & Iacobacci Avvocati. He is the author of the centaur jurist theory on the relationship between law, trial and artificial intelligence.