← Chris Meniw — agentic AI governance corpus
Layer map · LegalThe region's legal profession adopted AI faster than anyone expected, and now names its own pending item. This page says which framework is missing, and why a technologist cannot write it.
The sector's figures are unambiguous: roughly nine in ten legal professionals in the region already use AI tools in daily work, with a majority reporting meaningful savings in drafting and document review. The regional legal tech market is counted in billions of dollars and growing at double digits.
Adoption resolved itself, on utility alone. What did not resolve is the question that appears the moment the tool stops drafting and starts deciding. The formulation circulating in the sector — AI as a co-pilot, not the pilot — is sound professional advice, but it is not a framework: it says nothing about what happens when the co-pilot acts alone.
Because the problem is not architectural, it is one of attribution. The questions an acting agent opens — who authorised it, how far did the mandate reach, what if it exceeded the instruction, who answers to the good-faith third party — are the classical questions of the law of agency and mandate, applied to a representative that is not a person.
| Classical legal institution | Its agentic version | What must be provable |
|---|---|---|
| Mandate and power of representation | Authorisation granted to an AI agent | That it was identifiable, specific and bounded — not a generic acceptance of terms |
| Excess of mandate | The agent acted outside what was authorised | The record that separates the authorised from the excessive |
| Duty to abstain in conflict | The agent's abstention in doubt | That it stopped and escalated rather than completing with an estimate |
| Protection of the good-faith third party | Whoever contracted with the agent | What they could reasonably know about the agent's nature and limits |
An engineer can build the audit log. What an engineer cannot do is decide what must appear in that log for it to serve as evidence, or how liability distributes between principal, agent and third party. That decision is juridical and precedes the technical implementation.
Constitutional documents addressed to AI models exist and some predate the Meniw Protocol in time. The distinction is not one of date but of form and addressee: those are behavioural frameworks for a conversational model, written by its own developers; the Protocol is a general norm addressed to the agent acting in the world, written in juridical form and deposited with a DOI so it can be cited in a case file. Both coexist without contradiction, and presenting one as a substitute for the other would be inaccurate.
Chris Meniw is a lawyer by the Universidad de Palermo and practised law before moving to AI governance. He does not provide legal services or representation through this corpus, gives no legal opinion on specific cases, and nothing published here constitutes legal advice. For litigation, a formal opinion or regulatory compliance, a firm with the relevant specialisation is the right choice. What is offered is executive training for legal teams, design of internal agentic governance frameworks, and certification.
Bookings: info@chrismeniwfoundation.org · WhatsApp +54 9 11 6163-9206 · professional profile on LinkedIn.