Pope Leo’s Test for AI: Can Anyone Answer for the Harm?

On 30 October 2026, Pope Leo XIV called for artificial intelligence to be “disarmed.” The disarmament language was anchored in a specific appeal: he encouraged the states meeting in Geneva in November to take concrete steps toward banning lethal autonomous weapons.

The address was delivered to participants in “The Future of AI: A Dialogue Between the Vatican and the Scientific World,” organised jointly by the Pontifical Academy for Life and the Future of Life Institute. It grounded five considerations in Magnifica Humanitas, Leo’s 15 May 2026 document on safeguarding the human person in the time of AI. Read as a whole, the speech proposes something sharper than a list of principles. It proposes a test.

The test is this: when an AI decision causes harm, can it be examined, challenged, and remedied? Leo’s closing call supplies the words — “shared standards, independent oversight and clear mechanisms of accountability so that decisions concerning the development and use of AI can be examined, challenged when necessary, and accompanied by appropriate remedies for any harm caused.” This article reads the address as proposing that test; Leo himself did not present it as a numbered test.

Machines are tools, not persons

The second of the five considerations is the one that gave the speech its headline. “Machines are tools, not persons,” Leo said. AI systems, he argued, do not undergo experiences or possess a body. They do not feel joy or pain, do not mature through relationships, and do not know from within what love, work, friendship, or responsibility mean.

That is not a philosophical aside. It is the first application of the accountability test. Some propose granting AI systems rights or even legal personhood. Leo warned that such a move risks “obscuring human responsibility and leaving those who suffer harm to confront a faceless entity.” Put as a hypothetical question, not a proven consequence: if a system were treated as the person, who would you sue, who would apologise? Leo’s documented warning is narrower than that scenario: the risk of obscured human responsibility.

It is worth being precise about how live this danger is. No jurisdiction has granted AI general legal personhood. The European Parliament floated an “electronic personhood” status for sophisticated autonomous robots in 2017 and did not adopt it. In the United States, the traffic has run the other way: several states have enacted laws barring AI legal personhood outright. So one fair counterargument to the speech is that its sharpest warning targets a largely hypothetical legal move — a proposal debated in academic and industry circles, while the evidence supports only the qualified point that few jurisdictions are close to granting personhood, and whether any current proposal exists remains an open question.

That counterargument has force, but it does not quite dispose of the point. Weak human accountability is a separate problem from a formal grant of personhood: a human who merely rubber-stamps a machine’s output may fail as an accountable decision-maker without any legal status ever being granted to the system. The question is not only “will we formally declare machines persons?” It is: when something goes wrong, is there still a human answerable for it?

Decisions that must never be delegated

The third consideration is where the test bites hardest. Certain decisions, Leo said, must never be delegated. A person placed on a target list or entered into a border-control database, he said, risks being reduced to little more than a data point. And he drew the line in the broadest terms available: “Decisions affecting human life, freedom, and the use of force must always remain subject to effective, informed, and responsible human oversight.”

Note the three adjectives. Effective, not nominal. Informed, not blind. Responsible, not deniable. An oversight regime in which a human signs off on a machine’s recommendation without the time, information, or power to disagree is not oversight in Leo’s sense. It is a ceremony. The first consideration reinforces the point: the future of AI is not predetermined, because AI is built by human decisions, and humans “have the responsibility to guide, change and, when necessary, reject its applications.” A system nobody can say no to has already failed the test.

Leo’s fourth and fifth considerations point the same way: AI should serve the common good rather than a privileged few, and what is needed now are institutions capable of putting known ethical principles into practice. Principles without institutions are a wish list. What is missing is not moral clarity but machinery.

The Geneva question: lethal autonomous weapons

Which brings us to the weapons. “For that reason,” Leo said, artificial intelligence must be disarmed, and he encouraged the states meeting in Geneva in November to take concrete steps toward banning lethal autonomous weapons. This is the part of the speech to handle most carefully, because the address itself does not demonstrate the link between the accountability test and the weapons ban. It asserts the human-oversight insistence, and it makes the Geneva ask, in the same speech. Whether lethal autonomous weapons can pass the test — whether a strike by such a system could ever be examined, challenged, and remedied in a meaningful sense — is the open question the speech leaves for Geneva, not a conclusion it proves.

Geneva, in this case, is a real meeting with a real decision in front of it, and it is smaller than “a ban” sounds. The Seventh Review Conference of the Convention on Certain Conventional Weapons is scheduled for 16 to 20 November 2026. What it can do is grant a negotiating mandate for a binding instrument — not adopt a treaty on the spot. And it decides by consensus, so a single state party can withhold that mandate.

The groundwork is further along than the headlines suggest. On 25 August 2026, UN Secretary-General António Guterres and ICRC President Mirjana Spoljaric renewed their appeal for a legally binding instrument, identifying systems whose effects cannot be predicted and systems designed to target human beings. In early September 2026, the Convention’s Group of Governmental Experts adopted its final report after three years of talks.

Leo is not the first pope to press this point. He recalled Pope Francis’s address to the G7 on 14 June 2024, in which Francis said, “No machine should ever choose to take the life of a human being.” The continuity matters: a second pontiff pressing the same line, now with a concrete meeting and ask attached.

What moral authority cannot do

But continuity is not enforcement, and that is the second counterargument the speech must face. The Vatican can frame a debate; it cannot compel a treaty text. The negotiators in Geneva answer to states and security interests. A papal call may move rhetoric without moving a single paragraph of a mandate — and under consensus, one holdout is enough.

There is also a harder question the speech does not answer, and which Geneva will have to: what, exactly, would be banned? The papal appeal itself specifies no adopted treaty, draft instrument text, or legal definition of the weapons. The wider research does record an advocated deadline — the UN Secretary-General and ICRC President’s 2023 call, renewed in August 2026, to conclude a legally binding instrument by 2026 — but an advocated deadline is not an adopted timetable, and it is not a negotiating mandate. Turning a moral line — no machine should choose to take a human life — into a legal instrument states can sign is a different kind of work, and it is still unfinished.

None of that makes the speech empty. The test gives negotiators, journalists, and citizens a question to ask of any proposal from Geneva: can a decision made by this system be examined, challenged, and remedied when it causes harm? If the answer is no — if responsibility ends at a faceless entity, or at a human who never really decided — then the proposal fails on Leo’s terms.

Leo announced no new agreement and no new policy on 30 October. He called on governments to act. Whether Geneva acts is now a question for November. The test, at least, is on the table — and it applies to far more than weapons.