VIGÍA
A pipeline to make AI risks visible in official discourse and accelerate AI Governance in Mexico.
Team
The people behind VIGÍA
-
Pilar Carolina Moncada Santana
Co-founder · Platform conception
-
Jason Maximiliano Pinelo Hau
Co-founder · Platform conception
A summary of the article “VIGIA: A Pipeline to Accelerate AI Governance in Mexico”, by Pilar Moncada and Max Pinelo.
The problem
What is not mentioned does not exist
Mexico uses frontier AI but does not produce it, and an AI safety vocabulary barely exists in the country. As a result, AI risks are not named in official discourse and remain invisible to those who could drive their regulation.
The article calls this invisibilization a higher-order risk: it causes no harm by itself, but it keeps first-order harms, such as failures, discrimination or surveillance, from being noticed and corrected. It also hides itself: a document with no mention of bias or human oversight does not look incomplete, it looks normal.
It is not the only bottleneck for AI Governance in Mexico, but it is an important one, upstream of several of the others.
The proposal
A six-step pipeline
To clear that bottleneck, the pipeline analyzes official and public discourse and makes AI risks visible.
- Identify the documents where the State publishes its AI initiatives and uses.
- Find the mentions of AI use in them, explicit and implicit.
- Classify the risk associated with each mention into 15 categories.
- Report each finding: why it is a risk, how urgent it is, and initial recommendations.
- Prepare reports for researchers, legislators and the general public.
- Measure how far the AI risk lexicon is adopted in public discourse, for example in digital newspapers.
The implementation
VIGÍA analyzes Mexico's official gazette
VIGÍA (Valoración de Impactos Gubernamentales de IA, Assessment of Governmental Impacts of AI) is the first implementation of the pipeline, built for Mexico. It uses the Diario Oficial de la Federación (DOF), Mexico's official gazette, because that is where the acts of the Mexican State acquire legal validity: a use of AI published in the DOF is no longer a plan but a fact of government with a date and a signature. It is also public and downloadable, so the exercise can be reproduced.
A language model (Claude Opus 4.8) read 153 complete issues of the morning edition, from 3 November 2025 to 12 June 2026. For each mention of AI use it records the original fragment, its location, the risk categories with their severity and one or more recommendations. Steps 5 and 6 are outside the scope of this first iteration.
Results
The risk lexicon is absent
Of 18 terms from the AI risk vocabulary, 16 do not appear even once in the 153 issues. The three cases that do appear come from two public research centers and a bibliographic citation, none of them a regulator.
When AI appears, it is almost always as a capability or benefit (72.7% of mentions) or as a purchase line item (11.7%). Only 18 mentions treat it as a source of harm, and all of them borrow the framing of an earlier legal regime, such as copyright or academic integrity.
Of the model's 665 findings, 636 (95.6%) flag a regulatory gap. The most frequent risks are privacy invasion, discrimination and bias, failures and errors, and authoritarian surveillance, and 50 of the 153 issues were marked as requiring urgent attention.
One example: on 29 May 2026 the Technical Committee of the Coatlicue supercomputer was created, the most explicitly AI-oriented instrument in the corpus. It contains no provision on permitted uses, model evaluation or access.
What comes next
Limitations and next steps
This is a first version. A single language model did all the analysis, with no second model to compare against, and the documents are enormous: the median DOF issue is around 389 pages. The risk categories also come from frameworks of countries that produce frontier AI and may not capture everything that matters in a country like Mexico.
The next steps are to validate the classification with a second model and human reviewers, prepare reports for those who drive AI Governance, and measure how far the risk lexicon circulates in public discourse. The pipeline can be replicated in other countries that use frontier AI without producing it, and in the future it can support a defense strategy for Mexico through regulation.
Explore the observatory
The results of this first implementation are published in the VIGÍA app.
Open the VIGÍA app