AI ObservatoryCosta Rica

Analysis

Costa Rica fell 17 points behind Chile in ILIA 2025

Regional benchmarking, structural capability gaps and state of the regulatory framework. Verified data from official sources and multilateral reports.

State & Algorithm

Biweekly series of analyses on AI in the Costa Rican state. Each issue unpacks a piece of the inventory: what works, what is stalled, and the open questions.

Regional benchmark

ILIA score, investment and executing body for the five Latin American countries in the 2025 ranking.

CountryILIA 2025AI investmentExecuting bodyKey milestone
Chile70.56USD 32MCENIA (National AI Center), operationalLatam-GPT, regional 70B-parameter model built for USD 550K.
Brazil67.39USD 4,000M (PBIA)Brazilian AI Plan (PBIA), led by MCTILargest absolute AI investment in Latin America.
Uruguay62.32Agesic with operating budget since 2007AgesicNational AI Strategy 2024-2030 approved with 40+ institutions; Ceibal program trained 80,000 students in AI. In December 2025, Decree 276/025 established the regulatory sandbox framework for AI and data spaces (supervised controlled testing environments). In April 2026, Parliament and Ceibal launched an AI training cycle for legislators and public officials. ↗1 ↗2 ↗3
Colombia55.84National AI Policy approved (CONPES)MinTIC + DNPAI CONPES approved and active digital transformation agenda.
Costa Rica53.83No explicit AI budget in national budgetMICITT (National AI Center of Excellence pending)ENIA 2024-2027 published (first national AI policy in Central America). Hacienda recovered ₡8,000M with AI in 2025.

Seven capabilities Costa Rica has not yet built

Compared with Estonia (1.3M people) and Singapore (5.6M people), two global benchmarks in digital government. Each gap cites verifiable evidence.

National interoperability platform (X-Road style)

Reference: Estonia

X-tee/X-Road in Estonia connects around 1,500 institutions and companies (about 600 public bodies), interoperates ~1,600 information systems and recorded close to 600 million queries last year.

Status in Costa Rica

No equivalent national interoperability layer exists. Each institution integrates at its own pace and on its own standards.

Why it matters

Without this layer, scaling pilots like EDUS or Hacienda Digital to other institutions means building each connection from scratch. It is the largest structural bottleneck preventing AI in government from moving past isolated cases.

↗ Open

Formal AI governance framework

Reference: Singapore

In 2024 Singapore published a Model AI Governance Framework for GenAI with 9 dimensions (accountability, data, trusted development, incident reporting, value alignment, model evaluation, transparency, security and AI for public good).

Status in Costa Rica

Costa Rica has five bills in progress (23.771, 23.919, 24.484, 24.875, 25.171), two with committee reports, none approved. The only operating framework is the Judicial Branch GenAI Guidelines, sectoral and not national.

Why it matters

Without a formal framework, each institution decides on its own how to evaluate risk, transparency and accountability. The uncertainty slows adoption and makes external audit difficult.

↗ Open

Unified national citizen chatbot

Reference: Estonia

Bürokratt is Estonia’s national conversational assistant, deployed across selected agencies and running 24/7 with voice, chat, SMS, email and sign language. Recognized by UNESCO/IRCAI as one of the Top 100 global AI projects in 2024.

Status in Costa Rica

There is no national citizen chatbot. The Hacienda Virtual Assistant is the only operational case in the central government, limited to its own domain.

Why it matters

A unified assistant reduces channel fragmentation and lowers the citizen’s entry barrier to the digital State. It is one of the highest visible-impact investments per dollar.

↗ Open

Internal AI assistant for public servants

Reference: Singapore

Pair, GovTech Singapore’s internal AI assistant, has 11,000+ users and 4,500+ weekly active users across the public administration.

Status in Costa Rica

No equivalent initiative is reported in the Costa Rican State. Any use of generative AI by public servants happens in commercial tools without institutional oversight.

Why it matters

Without an institutional tool, risk shifts to individual public servants: sensitive state data on external platforms, with no traceability. An internal tool also captures measurable productivity.

↗ Open

Mandatory AI testing before deployment

Reference: Singapore

GovTech Singapore runs Litmus (stress-testing for GenAI before deployment) and Sentinel (real-time content moderation) as a mandatory layer over government AI systems.

Status in Costa Rica

No national pre-deployment testing framework exists. Each institution decides what testing to apply before pushing an AI system to production.

Why it matters

It is the difference between an isolated AI failure and a public-trust crisis. A standard framework also lowers adoption cost for smaller institutions unable to build their own.

↗ Open

Numerical AI talent target with budget

Reference: Singapore

Singapore’s National AI Strategy 2.0 sets a target of tripling the AI talent pool to 15,000 people, with S$20M+ allocated to SG Digital scholarships and international internships.

Status in Costa Rica

ENIA 2024-2027 does not set a numerical AI talent target with associated budget. The +4,500 LINC participants and the 90,000 MEP teacher scholarships count as general digital literacy, not as a specific AI target.

Why it matters

Without a target and budget, there is no way to evaluate progress or hold anyone accountable. Any conversation about the talent gap stays purely qualitative.

↗ Open

Cross-border government data agreements

Reference: Estonia

Since 2018 Estonia and Finland operate the first cross-border interoperability federation based on X-Road, with bidirectional government data flows.

Status in Costa Rica

There is no federation of government data between Costa Rica and other countries in the region. Existing international integrations (Central America, OECD) do not include operational data flow between administrations.

Why it matters

For services that cross borders (migration, trade, regional health), the lack of formal agreements forces manual processes. It is a gap that prevents Costa Rica from leveraging the regional leadership it already holds in AI dialogue.

↗ Open

Five bills, zero laws passed

The five AI bills filed between 2023 and 2025 remain unapproved. Two already have a committee report. Costa Rica is moving forward without a formal regulatory framework.

File
23.771

Law to Regulate Artificial Intelligence in Costa Rica

General framework to regulate the development, use and application of AI in Costa Rica. Creates the AI Regulatory Authority (ARIA). Has a positive majority committee report (September 2024), but faces opposition from CAMTIC and other tech ecosystem stakeholders.

In committee
File
23.919

Law for the Responsible Promotion of Artificial Intelligence in Costa Rica

Creates controlled spaces for regulatory experimentation with emerging AI technologies. Modeled after financial sandboxes in the fintech sector. Has a substitute text approved in committee (March 2025).

Committee report issued
File
24.484

Law for the Implementation of Artificial Intelligence Systems

Proposes a tiered classification by risk level (similar to the EU AI Act). Has an approved substitute text (October 2025). The most recent and technically robust of the five bills.

Committee report issued
File
24.875

Law to Regulate the Use of Artificial Intelligence in Electoral Processes

Establishes disclosure obligations when AI-generated content is used, prohibits electoral deepfakes, automated fake accounts, and micro-targeting strategies designed to undermine electoral integrity. Fines range from 10 to 50 base salaries. Filed by Vanessa De Paul Castro Mora (PUSC).

In committee
File
25.171

Law for the Protection of Personal Traits and Artistic Creations in Digital Environments

Criminalizes identity impersonation through digital means, including unauthorized use of voice, image, and artistic creations. Aimed at combating deepfakes and digital identity fraud, rather than regulating AI as such.

In committee

This analysis presents evidence and gaps. It does not include public-policy recommendations: that conversation belongs to the country’s institutional actors.