Direct comparison

Latin America AI Law: 5 Countries Compared

Only Peru has an enacted AI law in Latin America. Mexico, Argentina, Chile and Colombia are still at bills, guidance or policy stage — compared side by side.

Written and maintained by CASRAI Editorial Board

Last updated 25 September 2026

Ask CASRAI · free to try

Ask about Latin America AI Law: 5 Countries Compared

Ask your first 2 questions free below. Subscribers get 150 a day for $29 a month.

An AI assistant specialized in research administration. It cites the sources behind every answer, labels web answers and says when it can't answer.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Ask your first question free Subscribe — $29/month See what it covers →Already subscribed? Sign in

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays free to read, with no account and no card.

In summary

CASRAI's frontier-ai-safety coverage of the Americas can create a false impression by proximity: readers who have seen guides on the EU AI Act, US state laws, Canada's failed AIDA and Brazil's stalled PL 2338/2023 can reasonably assume the rest of the hemisphere is close behind. It isn't, and the gap is wider than "still drafting." Of five major Latin American jurisdictions, only Peru has an enacted, in-force national AI statute — Ley N° 31814, published 5 July 2023 — and even that law is a short, principles-only, promotional text with no risk tiers and no penalties of its own; the real risk-classification regime that makes Peru resemble the EU AI Act came two years later, via Decreto Supremo N° 115-2025-PCM (published 9 September 2025), whose general provisions took effect only in mid-January 2026. Mexico has multiple federal AI bills and an active academic-industry policy debate but no enacted statute; as of a September 2026 policy panel covered by Infobae, participants were still debating what a future reform should even regulate. Argentina's national government issued non-binding 'Recomendaciones para una Inteligencia Artificial Fiable' in December 2023 — guidance, not law. Chile's Boletín 16821-19, 'Regula los sistemas de inteligencia artificial,' has been before Congress since 7 May 2024 and remains 'En tramitación' (in process) with no enactment date. Colombia has no enacted AI statute either; its most recent instrument, CONPES 4144 (2025), is a national policy document, built on a 2021 Ethical Framework for AI that is soft-law guidance for public entities, not a source of legal obligations. The table below sets all five side by side on legal status, binding force, penalties, scope and effective date, so the one real exception (Peru) isn't mistaken for a regional pattern.

How do Peru — Ley 31814 (2023) + D.S. 115-2025-PCM, Mexico — Federal AI Bills (unenacted), Argentina — 'Recommendations for Reliable AI' (2023), Chile — Boletín 16821-19 (pending), Colombia — CONPES 4144 (2025) + 2021 Ethical Framework compare side by side?

The table below compares Peru — Ley 31814 (2023) + D.S. 115-2025-PCM, Mexico — Federal AI Bills (unenacted), Argentina — 'Recommendations for Reliable AI' (2023), Chile — Boletín 16821-19 (pending), Colombia — CONPES 4144 (2025) + 2021 Ethical Framework across 7 procurement-relevant dimensions, from legal status through what would change this comparison.

Side-by-side comparison

Dimension	Peru — Ley 31814 (2023) + D.S. 115-2025-PCM	Mexico — Federal AI Bills (unenacted)	Argentina — 'Recommendations for Reliable AI' (2023)	Chile — Boletín 16821-19 (pending)	Colombia — CONPES 4144 (2025) + 2021 Ethical Framework
Legal status	Enacted. Ley N° 31814 was passed by Congress and published in <em>El Peruano</em> on 5 July 2023 — the first enacted national AI statute in Latin America. Its implementing regulation, D.S. N° 115-2025-PCM, was published 9 September 2025.	Not enacted. Multiple federal AI bills have been introduced in Congress over several sessions; none has passed. As of a September 2026 academic-industry policy panel (reported by Infobae), participants were still debating scope and approach for a future reform, not implementing one.	Not a law. 'Recomendaciones para una Inteligencia Artificial Fiable' (Recommendations for Reliable AI) was issued by Argentina's national government in December 2023 as non-binding guidance, not legislation.	Bill pending. Boletín 16821-19, 'Regula los sistemas de inteligencia artificial,' was introduced 7 May 2024 and remains listed 'En tramitación' (in process) in the Senate's own bill-tracking system, with no enactment date.	Not an enacted statute. CONPES 4144 (2025) is a National Council for Economic and Social Policy document — a policy instrument, not legislation — building on a 2021 Ethical Framework for AI that is soft-law guidance for public-sector AI projects.
Binding force	Mixed by document. The 2023 law itself creates no binding obligations — it is promotional and principles-only. The 2025 decree does create binding obligations for systems it classifies as high-risk (registry, security protocols, human oversight, transparency duties).	None. No bill has been enacted, so there is no binding federal AI-specific obligation in force.	None. A recommendations document has no binding force; it does not create enforceable duties for public or private actors.	None yet. A bill in Congress has no legal force until enacted; whatever the final text requires depends on what survives the tramitación process.	None. A CONPES policy document and an ethical framework are both non-binding by design — they guide public-sector practice and signal policy direction without creating enforceable legal duties.
Penalties for non-compliance	No AI-specific penalty regime in either document. The 2025 decree gives Peru's digital-government secretariat (SGTD) a monitoring and referral role, not independent sanctioning power — violations are referred to Peru's existing data-protection, consumer-protection, cybercrime and public-sector oversight regimes, which enforce under their own pre-existing rules.	Not applicable — no enacted law to attach a penalty to.	Not applicable — a recommendations document carries no penalty provision.	Not yet determined. Boletín 16821-19's enforcement and penalty provisions, if any survive committee review, are not settled while the bill remains in tramitación.	Not applicable — CONPES documents and ethical frameworks do not carry penalty clauses; Colombia's Constitutional Court has addressed algorithmic transparency in specific rulings (e.g. T-067/25), but that is case law on existing rights, not a penalty regime created by an AI statute.
Scope / who it reaches	Broad in intent (any AI use touching Peru's digital-transformation goals) and specific in the 2025 decree, which names eight high-risk categories: critical infrastructure, employment decisions, consumer credit, healthcare access, education assessment of minors, social-programme access, workplace emotion inference, plus an open discrimination-risk clause.	Undetermined — no single bill has consolidated into an enacted scope; proposals under discussion range across sector-specific ideas (health, education, financial services, intellectual property) rather than one settled framework.	General — the recommendations address AI development and use broadly, aimed at public and private actors, but as guidance rather than a scoped legal obligation.	As proposed: AI systems generally, following the now-common risk-tiered approach (the bill's title, 'Regula los sistemas de inteligencia artificial,' signals a general AI-systems law) — final scope depends on what Congress passes, if anything.	CONPES 4144 sets national policy direction across sectors; the 2021 Ethical Framework specifically targets public entities' AI projects, not private-sector AI generally.
Effective date	Law: 5 July 2023 (date of publication). Decree's general provisions: around mid-January 2026 (90 business days after its 9 September 2025 publication); sector-specific compliance windows run one to four years from September 2025.	None — not enacted.	Issued (not "effective" in a legal-force sense) December 2023.	None — still in Congress as of this comparison; no enactment or effective date exists until it passes.	CONPES 4144 approved/published 2025 as policy; the 2021 Ethical Framework has applied as guidance since 2021. Neither carries a legal "effective date" in the statutory sense.
Risk-classification approach	Three-tier system under the 2025 decree: prohibited uses (Art. 23), high-risk categories (Art. 24), and a general acceptable-risk tier (Art. 22.2) — structurally similar to the EU AI Act's pyramid, though without the EU's independent conformity-assessment or fine machinery behind it.	No consolidated approach yet; discussion has touched on differentiating "risk" systems from those eligible for lighter-touch treatment, but nothing has been settled in an enacted text.	Not risk-tiered — the recommendations set out principles for trustworthy/reliable AI rather than a classification scheme with tiers attached.	Expected to follow a risk-tiered model based on the bill's framing, consistent with the regional and EU trend, but the specific tiers and thresholds are not settled while the bill remains in committee.	Not risk-tiered — CONPES 4144 and the 2021 Ethical Framework set policy direction and ethical principles rather than a binding risk classification.
What would change this comparison	Already the regional exception; the open question is enforcement — whether SGTD's referral-only model gets a dedicated AI sanctions power in a future amendment.	Enactment of a consolidated federal bill following the kind of scope debate covered in the September 2026 policy panel — sector carve-outs (health, education, finance, IP) were explicitly discussed as one possible path.	Congress passing binding legislation building on the 2023 recommendations, rather than further non-binding guidance.	Boletín 16821-19 clearing committee and both chambers — the single biggest legislative move that would add a second enacted AI law to the region.	Congress enacting a binding AI statute grounded in the CONPES 4144 policy direction, rather than further CONPES documents or ethical-framework updates.

Common questions

Common questions about Peru — Ley 31814 (2023) + D.S. 115-2025-PCM vs Mexico — Federal AI Bills (unenacted) vs Argentina — 'Recommendations for Reliable AI' (2023) vs Chile — Boletín 16821-19 (pending) vs Colombia — CONPES 4144 (2025) + 2021 Ethical Framework

Is Peru really the only Latin American country with an enacted AI law?

Among these five major jurisdictions, yes — and it's not close. Mexico, Argentina, Chile and Colombia each have real activity (bills, recommendations, policy documents) but none has passed an enacted, AI-specific statute. Brazil, covered separately in CASRAI's guide on <a href="/guides/brazil-ai-bill-pl-2338-status/">PL 2338/2023's status</a>, is the other near-miss in the region: its bill was approved by the Senate but remains stalled in the Chamber of Deputies, unenacted. Peru earned its exception twice — Ley 31814 in 2023, and its implementing decree in 2025 — which is why CASRAI covers Peru's law in its own dedicated depth rather than only as a row in this table.

Doesn't Peru's 2023 law actually do very little on its own?

Correct, and that nuance is easy to lose in a one-line "Peru has an AI law" summary. Ley 31814 is promotional: it sets principles, names a governing authority (SGTD), and directs the executive branch to build out substantive rules later. It does not itself classify AI by risk or create penalties. The risk tiers that make Peru resemble the EU AI Act — prohibited uses, high-risk categories, concrete obligations — arrived two years later, in Decreto Supremo N° 115-2025-PCM. A reader citing only the 2023 law and expecting EU-AI-Act-style compliance exposure would be wrong in one direction; a reader who thinks Peru still has no real rules, post-2025, would be wrong in the other. See CASRAI's full breakdown in <a href="/guides/peru-ai-law-31814-explained/">Peru's Ley 31814: A Promotional Law, Then a Real Risk Regime by Decree</a>.

Why does Chile's bill matter if it hasn't passed?

Because "in Congress" and "not happening" are different things, and Boletín 16821-19 is real, active legislative activity — introduced 7 May 2024, still moving through committee as of this comparison. It's the most likely of the four pending jurisdictions to become the region's second enacted AI law, precisely because it already exists as a specific, numbered bill with a title ("Regula los sistemas de inteligencia artificial") rather than a strategy document or a set of recommendations. Its final scope, tiers and any penalty provisions are not settled until it clears both chambers.

What's the practical difference between Colombia's CONPES 4144 and an actual AI law?

CONPES documents are Colombia's mechanism for coordinating national economic and social policy across ministries and agencies — they set direction, targets and (sometimes) budget, but they do not create legal obligations enforceable against private parties the way a statute does. CONPES 4144 (2025) sits alongside a 2021 Ethical Framework for AI that guides public entities' own AI projects, plus a 2024 Judiciary Council AI guideline and a 2025 Constitutional Court ruling (T-067/25) on algorithmic transparency — a genuinely active policy landscape, but one built from policy instruments and case law, not an enacted AI statute.

Does NIKOLAI have anything to say about comparing these regimes?

Nothing specific to this comparison beyond a general caution. NIKOLAI — CASRAI's own independent, unendorsed frontier-AI-safety dictionary — defines a <a href="/nikolai/element/mapping-declaration/">Mapping Declaration</a> as a statement filed by a regime's own authority that its terms correspond to another regime's, which is a stronger claim than an outside analyst's reading. None of Peru, Mexico, Argentina, Chile or Colombia has filed such a declaration with NIKOLAI, and the parallels drawn on this page — describing Peru's risk tiers as resembling the EU AI Act's, for instance — are CASRAI's own <strong>shadow mappings</strong>, not official equivalences endorsed by any government named here.

Going deeper