The Pentagon runs La Tilde, a Spanish-language news site aimed at Latin America that produces content with generative AI at scale, with no bylines, no visible masthead and with its only disclaimer (“funded by the US government”) buried on the About page. The Intercept revealed on June 2 that the operator is SOCSOUTH (US Special Operations Command South), the same unit that in April 2026 ran similar operations in Arabic for the Middle East.
What makes the case unique is not just the propaganda itself (that is nothing new) but its economics: the marginal cost of producing geotargeted content with language models approaches zero. La Tilde’s editorial mix is deliberate: personal finance guides sit alongside articles praising US military operations in the region. The most powerful actor in the world has already figured out how to direct persuasive content to millions of Spanish-language readers for a practically nonexistent price. The question left on the table: what tools and what budget do fact-checking outlets and LatAm governments have to detect it in real time?
Also today
-
“Great American AI Act”: bipartisan bill freezes state AI laws for 3 years — A 269-page discussion draft presented in the US Congress would suspend for three years state rules that regulate the development of AI models, halting the Colorado AI Act and California’s rules, the main references for lawmakers in Colombia, Chile and Argentina when they design their own laws.
-
OpenAI Codex surpasses 5 million weekly users and launches six role-based plugins — 20% of users are already non-technical professionals, and that group is growing three times faster than developers. The new Legal and Corporate Finance plugins, plus the Sites feature for creating web applications without writing code, consolidate the transition from a tool for programmers to a universal work platform.
-
Colombia, runoff: Espriella leads by 7 points according to AtlasIntel — The June 21 election arrives with an enforcement gap: Law 2502, which defines AI impersonation as an offense, was approved, but its penalties only take effect in July, ten days after the vote. Audio cloned for as little as $10 is circulating on WhatsApp.
-
OpenAI integrates GPT-5.5 into GPT-Rosalind for drug discovery and genomics — The system now covers the entire life sciences pipeline with agentic capabilities. Access is restricted to institutions with a current agreement with OpenAI, which excludes most universities in Latin America.
In the region
The “Great American AI Act” is the third global governance event with a direct impact on LatAm in less than a week (on top of the EU Tech Sovereignty Package on June 3 and Trump’s executive order on voluntary AI on June 2), and the region had no seat at any of those tables. If the proposal moves forward, it freezes precisely the regulatory references that lawmakers in Colombia, Chile and Argentina have been adapting in their own bills. In parallel, the new OECD report Governing with Artificial Intelligence (101 pages with more than 200 use cases of AI in government, published on June 3) includes a chapter on public procurement of AI directly applicable to ChileCompra, SECOP (Colombia) and ComprasNet (Brazil).
Launches
-
LATAM Airlines’ AI travel agent — Available starting today in Colombia, Peru and Ecuador (it has been live in Chile since April). Built with Gemini and Vertex AI by a team of ten engineers; it has already accumulated more than 52,000 users in the Chilean beta. According to LATAM’s own data, 53% of travelers in the region already use AI to plan their trips.
-
OpenAI Codex — role-based plugins — Legal and Corporate Finance plugins rolling out to ChatGPT Plus subscribers or higher. The Sites feature generates interactive web applications that can be shared by URL without needing to write code.
-
Grok Voice — Spoken interaction with Grok available starting today on grok.com, iOS and Android (requires an xAI account). Spanish support has not been officially confirmed.
Threads we’re following
Since May we have been following how Colombia’s June 21 presidential runoff became the year’s sharpest laboratory case of regulatory timing failure: Law 2502, which punishes AI impersonation, exists on paper, but its penalties only take effect in July, exactly ten days after the vote. This week AtlasIntel puts Espriella 7 points ahead of Cepeda in the latest polls, and the window of technological impunity remains open for cloned audio distributed via WhatsApp, the cheapest and most credible risk vector identified by Colombian analysts.
If the cost of producing geotargeted propaganda at scale is practically zero, can fact-checking remain a handcrafted job funded by donors, or does it need to become public infrastructure?
About this entry. It is generated automatically from public sources, without human review before publication. It may contain errors of interpretation or summary; please check each story against its original source (the links lead there) before citing it or making decisions based on it.
Doble Click is written with Anthropic models.