On June 4, Jack Clark and Marina Favaro of the Anthropic Institute published “When AI Builds Itself”: a call to create a verifiable global mechanism that would allow labs to pause in a coordinated way if AI reaches the ability to improve itself recursively. The central argument for urgency is concrete: Claude already writes more than 80% of the code in Anthropic’s own codebase, and its engineers merge eight times more code per day than two years ago.
The proposal comes with context that deserves attention. Four days earlier, Anthropic filed a confidential S-1 with the SEC, and eight days earlier it closed its $65 billion Series H at a valuation of 965 billion. The structural conflict of interest is hard to ignore: the company that built one of the most capable AI systems in the world is asking everyone to stop at the same time, but only if everyone stops. For Latin America, the problem runs deeper: if such an agreement were ever negotiated, the region was not present in Glasswing, has no representation in CAISI, and was not consulted in any of last week’s three governance events (Trump’s executive order, the European package and the Great American AI Act).
Also today
-
OpenAI launches Dreaming V3: ChatGPT synthesizes your memory profile in the background — Factual recall rises from 41.5% to 82.8% according to the company itself. The European Data Protection Board (EDPB) warns that this constitutes profiling under the GDPR. The feature is active while ChatGPT is already running conversational ads in Brazil and Mexico.
-
Great American AI Act: bipartisan proposal freezes state AI regulation for 3 years — 269 pages that would block state laws on AI development in the US for three years. It closes a week with three global governance events in 72 hours: Trump’s executive order, the European package and this bipartisan bill.
-
LATAM Airlines expands its AI travel agent to Colombia, Peru and Ecuador — More than 52,000 active users in Chile since its launch. The agent suggests destinations, organizes itineraries and quotes flights with LATAM Pass miles. Built on Google Gemini and Vertex AI, available in the iOS/Android app.
-
Google pledges to replenish the water its data centers consume by 2030 — 19 billion gallons a year across 97 watersheds, plus replacing water cooling with air cooling at sites with vulnerable water resources.
In the region
Colombia enters the final stretch of its presidential campaign with the most fragile regulatory situation in the region: Law 2502, which defines AI impersonation as an offense, was approved, but its enforcement mechanisms only take effect in July, ten days after the June 21 vote. Without a body equivalent to the TSE with technical capacity in AI, and with voice cloning available for as little as $10 a month, the election arrives with active deepfakes already documented. According to AtlasIntel, Espriella leads with 50.3% against Cepeda’s 42.6%. It will be the first high-visibility test in the region of whether it is possible to defend an electoral process from generative disinformation without operational regulatory tools. At the global level, the week’s triple governance event consolidates two incompatible models: the European one, with obligations and data sovereignty; and the American one, with voluntary disclosures and state preemption. For Latin American regulators, the benchmark regulatory standard will remain the EU AI Act.
Launches
- LATAM Airlines AI Travel Agent — The first AI agent from a Latin American airline available in multiple countries; it suggests destinations, organizes itineraries and quotes flights with miles. Free in the iOS/Android app with a LATAM Pass account, in Spanish, now in Chile, Colombia, Peru and Ecuador.
Threads we’re following
Colombia and electoral deepfakes have been a developing story on this blog for weeks. We documented the concrete risk for the June 21 runoff and the regulatory gap of Law 2502; today that gap becomes more urgent: there are 15 days left until the vote and enforcement still has no operational tools. The election also becomes an indicator of something broader: if the countries in the region that already have approved laws cannot enforce them in real time, the debate about taking part in global AI governance agreements (like the one Anthropic proposes today) seems even more distant.
If AI already writes more than 80% of the code of those who build it, when do we stop talking about tools and start talking about actors? And if that transition has already happened, who in Latin America has the real capacity to evaluate it?
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.