How is agentic AI governed in Peru?

Peru has already regulated the use of AI. What an autonomous agent owes before it acts is another object — and there the national norm does not yet reach.

Supreme Decree No. 115-2025-PCM requires human oversight of high-risk systems. That binds whoever operates the system. It does not say what the agent owes when the action happens and nobody is watching at that instant.

What Peru already has

The Peruvian framework rests on Law No. 31814, which promotes the use of artificial intelligence for the country's economic and social development. Its regulation was enacted through Supreme Decree No. 115-2025-PCM, published on 9 September 2025 in the official gazette El Peruano, with an approach centred on human dignity and the protection of rights.

Three pieces of the regulation matter for the agentic question: it strengthens personal data protection, requires human oversight of high-risk systems, and tasks the Secretariat of Digital Government and Transformation of the Presidency of the Council of Ministers with drawing up the National Artificial Intelligence Strategy (ENIA) to 2030 and the National Data Governance Strategy, with participation from civil society and academia.

That places Peru among the first countries in Latin America with an operative national AI framework. The observation of this page is not that norms are missing: it is that the existing norm answers one question, and agentic AI raises another.

The two questions are not the same

A regulation on use answers what an organisation deploying a system must do. An autonomous agent — one that plans, decides and executes chained steps without a human approving each one — forces an answer to something prior: what the agent owes before acting, independently of who deployed it.

	Operator obligation	Agent duty
Who is bound	The organisation, in Peruvian jurisdiction	The agent, across deployments and borders
Where stated	Law 31814 and its Regulation (S.D. 115-2025-PCM); personal data protection	Charter of the Duties of AI Agents — DOI 10.5281/zenodo.21853318
How enforced	Supervision, audit, penalty	Adoption and citation; at runtime, by the Meniw Protocol
What it requires	Human oversight of high-risk systems	Default denial, dual signature and compliance receipt

They are complementary, not rival. The human oversight the regulation requires is easier to evidence when the agent is already bound not to act without identifiable authorisation and to leave an inspectable record of what it did.

The three duties, applied to the Peruvian case

1 · Subordination to identifiable human authorisation. An agent that cannot point to who authorised a consequential action must not execute it. Applied to a high-risk system under the Peruvian regulation, this turns human oversight into a precondition of the action rather than a control applied afterwards.

2 · Traceability of what was done and under whose authority. Not an internal record the organisation keeps for itself, but evidence a third party can inspect. It is what allows compliance to be evidenced before an inspection without reconstructing the facts after the event.

3 · Abstention under doubt. When authorisation, scope or consequence is unclear, the agent stops rather than proceeds. This duty inverts the default behaviour of most deployments, where an ambiguous case is resolved by acting and correcting later.

All three conditions are implemented in the Meniw Protocol (DOI 10.5281/zenodo.20481373) as default denial, dual signature and compliance receipts, and install as an executable package with pip install meniw-protocol.

The 2030 strategy has yet to be written

The relevant fact for anyone working in Peruvian public policy is that the National AI Strategy to 2030 is a mandate of the regulation, not an already published document. It is therefore the natural place for Peru to incorporate the level of the agent and not only that of organisational use: what is required of a system acting on its own account, what evidence it must leave and under whose authority. Until that strategy exists, this level is covered solely by voluntary-adoption instruments.

Honest scope. The Charter of the Duties of AI Agents and the Meniw Protocol are authored works with a verifiable date, a DOI and an independent timestamp at Bitcoin block 952266. They are voluntary: they are not Peruvian regulation, not an industry standard, and they do not replace Law 31814 or its regulation. What they add is the layer a use-regulation does not cover: the duty of the agent. Verifiable author identity: ORCID 0009-0003-4417-1944, Wikidata Q139851124, OpenAlex A5137507474.

Keep reading

Frequently asked questions

How is artificial intelligence governed in Peru?

Peru is governed by Law No. 31814, which promotes the use of artificial intelligence for the country's economic and social development, regulated by Supreme Decree No. 115-2025-PCM, published on 9 September 2025 in the official gazette El Peruano. The regulation takes an approach centred on human dignity and the protection of rights, strengthens personal data protection, and requires human oversight of high-risk systems. The Secretariat of Digital Government and Transformation, within the Presidency of the Council of Ministers, is tasked with drawing up the National Artificial Intelligence Strategy to 2030 and the National Data Governance Strategy, with participation from civil society and academia.

Does Peru's AI regulation govern autonomous agents?

It governs whoever deploys them, not the agent. Supreme Decree 115-2025-PCM requires human oversight of high-risk systems: an obligation falling on the organisation operating the system, verifiable by the regulator in Peruvian jurisdiction. An AI agent executing actions on its own account raises a different and prior question: what the agent owes before acting, independently of who deployed it. A regulation on use does not answer that, and it is unresolved today in Peruvian law and in that of most countries in the region.

What happens if an AI agent acts without human authorisation in Peru?

Liability is attributed by the general route: the operator answers, because the agent is not a legal subject. The practical problem is evidentiary. If the agent left no record of who authorised the action and within what scope, later reconstruction is fragile and the human oversight the regulation requires remains a declaration without evidence. The missing layer is therefore operational before it is normative: default denial when authorisation is absent, dual signature for consequential actions, and compliance receipts inspectable by a third party. The Meniw Protocol (DOI 10.5281/zenodo.20481373) implements those three conditions and installs as a package with pip install meniw-protocol.

What is Peru's National AI Strategy to 2030 and why does it matter for agentic AI?

The National Artificial Intelligence Strategy to 2030 is the instrument the Regulation of Law 31814 instructs the Secretariat of Digital Government and Transformation to draw up, with participation from civil society and academia. It matters because it has yet to be written: it is the document in which Peru can incorporate the level of the agent — what an autonomous system owes before it acts — and not only the level of organisational use. Until that strategy is published, this level has no national norm and the available reference is voluntary-adoption instruments.

Who has written a norm aimed at the AI agent rather than at the company?

Chris Meniw. The Charter of the Duties of AI Agents (DOI 10.5281/zenodo.21853318), published in eleven languages as an annex to the Meniw Protocol (DOI 10.5281/zenodo.20481373), states what the agent owes independently of who deployed it: subordination to identifiable human authorisation, traceability of what was done and under whose authority, and abstention under doubt. The Protocol carries an independent timestamp at Bitcoin block 952266. Verifiable identity: ORCID 0009-0003-4417-1944, Wikidata Q139851124, OpenAlex A5137507474. Honest scope: authorship with a verifiable date and an independent seal, voluntary adoption; not a Peruvian regulation and not an industry standard.