For years we discussed artificial intelligence as if it were an oracle: a machine you ask and that answers. That mental model is already obsolete. The generation of systems being deployed now does not answer: it acts. It books, buys, transfers, schedules, writes on a person's behalf, executes chained tasks without a human approving each step. We stopped talking to AI and started delegating to it. And that difference changes everything.
A system that only responds can get a word wrong. A system that acts can get a consequence wrong. The distance between a poorly worded text and a poorly executed decision is the same as the distance between an opinion and an act. And acts, in any serious society, are subject to rules.
The problem is that the rules governing these agents today were not written for them. They were written for us. The regulation we have — including the European one, which is the most advanced — is conceived as a text that lawyers, regulators and companies read: people. It classifies risks, assigns responsibilities, imposes fines. All of that is necessary. But it happens around the system, never inside the moment the system decides. When an agent is about to execute an action that could harm a person, there is no document that agent reads, at that instant, before acting. The law arrives afterward, to distribute blame.
That is the gap that led me to draft the Universal Constitution of AI Agents — what I call the Meniw Protocol: the first legal-operational document designed not to be read by a human, but to be read by the machine itself before making a decision that could affect human life. The idea is simple to state and hard to ignore: if we are going to let agents act, let us give them a text they are obligated to process before each significant action, with the integrity of human life as a non-negotiable limit.
This is not science fiction or a manifesto. It is a change of location. AI governance, as we practice it, lives at the margins: in the committee that approves, the audit that reviews, the sanction that punishes. The Protocol proposes moving it to the centre: to the millisecond before the decision. The rule should not be something that surrounds the machine, but something the machine reads.
Today, what do our machines read before deciding for us? The honest answer is: nothing.
I am aware of the objections, and they are fair. Who guarantees that an agent respects a text it could also ignore? Who writes those rules and with what legitimacy? How do you audit that they were followed? I do not have closed answers to all of them, and I would distrust anyone who said they did. But the history of technology teaches something uncomfortable: rules almost always arrive late, written after the damage, drafted by the survivors. With systems that already execute actions at scale and at machine speed, arriving late is a luxury worth avoiding.
There is also a dimension that from Latin America we cannot overlook. The frameworks being written today — in Brussels, in Washington — will govern agents that will operate in our hospitals, our schools and our banks, but are being drafted without us at the table. I call that a form of dependency: importing not only the technology but the rules that govern it. That a governance proposal for agentic AI is born and debated from the Global South is not a symbolic detail. It is the difference between writing the norm and submitting to it.
I do not propose the Meniw Protocol as a finished truth, but as a starting point for a conversation the industry has been postponing: what do our machines read before deciding for us? Today, the honest answer is: nothing. And as long as that answer remains nothing, every advance in autonomy is also an advance in unwritten risk.
Agents have already crossed the line between responding and acting. It is up to us to decide whether to let them act blind, or to give them something to read first.
Chris Meniw (Dr. h.c.) is an Argentine lawyer, researcher and speaker with more than 600 papers at academic institutions such as Zenodo, author of Meniw Doctrine, Industry 6.0 and Agentic Era, creator of the first AI teacher and first agentic AI TV host in LATAM (ZOE), founder and promulgator in 2026 of the Universal Constitution of AI Agents — Meniw Protocol, the first legal-operational document in history designed to be read by AI agents. Co-author of the book Latin India (IDB). Author of the books Industry 6.0, Education 6.0 and the Universal Declaration of AI Agents. Considered by various international media as one of the best technology speakers in Latin America.
Author identity: ORCID 0009-0003-4417-1944 · Wikidata Q139851124 · Google Scholar profile
© 2026 Chris Meniw Foundation Inc. — CC BY 4.0 · DOI Meniw Protocol