What happened
Latam‑GPT, billed as the first open‑source large language model created in and for Latin America, was released on 10 February 2026 by Chile’s National Center for Artificial Intelligence (CENIA). The initial version supports Spanish and Portuguese and incorporates a corpus of over 300 billion tokens. The project claims participation from roughly 200 professionals and more than 65 institutions across 15 countries. In August 2026, the University of São Paulo’s AI Center announced its involvement in the upcoming Latam‑GPT 2.0, confirming the inclusion of Quechua and indicating plans to add Brazilian indigenous languages such as Nheengatu and Bororo. CENIA has published ethical guidelines, licensing terms, and curation criteria, but the article notes that public documentation does not detail how indigenous authority over cultural knowledge will be exercised.
The National Center for Artificial Intelligence (CENIA) in Chile announced the launch of Latam‑GPT on 10 February 2026, positioning it as the first open‑source large language model (LLM) designed specifically for Latin America. According to CENIA, the model was trained on a curated dataset exceeding 300 billion tokens, primarily in Spanish and Portuguese, and was built with an emphasis on openness, equity, and cultural preservation.
CENIA reports that the development involved roughly 200 professionals and more than 65 institutions spanning 15 countries, suggesting a broad regional collaboration. The organization has released documentation outlining ethical guidelines, licensing arrangements, and data‑curation criteria, though the public record does not specify how indigenous communities’ rights over their cultural knowledge are protected or enforced.
In August 2026, the AI Center at the University of São Paulo (USP) confirmed its participation in the next iteration, Latam‑GPT 2.0. This version will add Quechua and is evaluating the inclusion of additional Brazilian indigenous languages such as Nheengatu and Bororo. The announcement underscores the growing focus on expanding linguistic coverage beyond the dominant colonial languages.
The article raises concerns that merely adding indigenous language data does not guarantee decolonial outcomes. It questions whether indigenous peoples will act as active participants in model design, data selection, licensing, and usage oversight, or remain passive data providers. The piece cites the CARE principles (Collective Benefit, Authority to Control, Responsibility, Ethics) and the Māori Te Hiku Media Kaitiakitanga licence as potential frameworks for ensuring community control.
Why it matters
Latam‑GPT represents a significant step toward regional AI autonomy, aiming to reduce reliance on models developed in North America or Europe. However, the model’s impact hinges on whether it can genuinely incorporate indigenous languages and knowledge in a decolonial, sovereign manner. Without clear mechanisms for free, prior, and informed consent, community control over data, and benefit‑sharing, the initiative risks reproducing extractive practices that marginalize the very cultures it seeks to serve. The debate highlights broader challenges for in multilingual, multicultural regions, where data sovereignty, cultural preservation, and ethical AI intersect. If successful, Latam‑GPT could provide a template for inclusive AI development; if not, it may reinforce existing power imbalances and undermine trust among indigenous communities.
Latam‑GPT’s regional focus challenges the current AI landscape, which is dominated by models from the United States, Europe, and China. By creating a model rooted in Latin American languages and contexts, the initiative could empower local developers, researchers, and enterprises with a tool better suited to regional nuances.
The model’s treatment of indigenous data is a litmus test for ethical in multilingual societies. If the project establishes robust consent mechanisms and community oversight, it could set a precedent for other regions grappling with similar issues of data sovereignty and cultural preservation.
Conversely, insufficient safeguards could perpetuate epistemic extractivism, where indigenous knowledge is harvested without adequate benefit‑sharing or control, eroding trust and potentially prompting legal challenges under international indigenous rights frameworks.
The model’s open‑source nature may encourage broader adoption, but it also raises questions about how downstream users will respect the original ethical commitments, especially when the model is fine‑tuned or commercialized by third parties.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Which component of an AI application is the machine-learning model itself?
What to watch next
Future releases of Latam‑GPT, especially version 2.0, will reveal whether indigenous participation moves beyond token data contributions to real decision‑making authority. Watch for the establishment of formal indigenous oversight bodies, consent protocols, and benefit‑sharing agreements. International observers will also monitor how the model’s licensing and open‑source status align with the CARE principles and whether independent audits verify compliance. The model’s adoption by governments, businesses, and civil‑society groups across Latin America will indicate its practical reach and the effectiveness of its governance framework.
The rollout of Latam‑GPT 2.0: language additions, governance structures, and any publicly disclosed consent processes.
Formation of indigenous advisory boards or decision‑making bodies that have formal authority over data usage and model outputs.
Independent audits or third‑party evaluations that assess compliance with CARE principles and any licensing conditions tied to indigenous content.
Adoption patterns by governments, NGOs, and private firms in Latin America, which will indicate the model’s practical impact and the effectiveness of its governance model.