On June 3, the European Commission published the most ambitious technological sovereignty package in its history: €420 billion for chips, data centers, cloud and artificial intelligence through 2036, along with a new law (the Cloud and AI Development Act) that classifies cloud providers into four sovereignty tiers and requires European public authorities to host sensitive data on the bloc’s infrastructure. The statement that sums up the stance: “We want to be sure nobody has a kill switch” (European Commission).
The same day, Donald Trump signed an executive order on AI that establishes voluntary 30-day access to frontier models (the most advanced on the market), with no prior obligations or advance review. It is the scaled-down version of an order his own administration had canceled in May under pressure from tech groups: the period went from 90 days to 30, the framework remains voluntary, and the vulnerability-sharing program includes Google, Microsoft and xAI, but not Anthropic. In less than 24 hours, the two blocs that shape the global AI ecosystem confirmed irreconcilable regulatory philosophies: the EU with mandatory sovereignty and layers of governance; the US with voluntarism and no prior controls.
For Latin America, which had no representation in either negotiation, this moment matters because its own regulatory frameworks (under construction in Colombia, Brazil, Mexico and Argentina) will tend to take one model or the other as a reference, often without explicit public discussion of which one they are adopting.
Also today
-
Microsoft presents 7 in-house MAI models at its Build 2026 conference — The family includes MAI-Thinking-1 (an MoE architecture comparable to the most capable reasoning models on the market) and MAI-Code-1-Flash (5 billion parameters, already available in VS Code and GitHub Copilot). Project Polaris will replace GPT-4 in Copilot starting in August: it is the first time Microsoft has decoupled its main products from OpenAI’s models at massive scale.
-
Meta launches Meta Business Agent globally on WhatsApp Business — An AI agent built into WhatsApp Business that can answer customers, schedule appointments and process sales; available globally with a free tier. Tiendanube is the first Latin American e-commerce platform integrated, and the product was already logging more than 10 million weekly conversations before the official launch.
-
The European package breaks down: €120B for chips, €200B for data centers, €100B for cloud and AI and €2B for open source — The Cloud and AI Development Act also includes an explicit open-source software strategy to reduce dependence on non-European providers in critical sectors.
In the region
Colombia enters its June 21 presidential runoff amid active use of AI in the campaign: voice cloning from platforms available for less than ten dollars is the main risk vector identified by media analysts. Law 2502, which provides for penalties for the use of electoral deepfakes, was approved, but its enforcement mechanisms will not take effect until July (ten days after the runoff), leaving the most sensitive electoral period without effective coverage. The contrast with the European-American debate is direct: while the big blocs define their frameworks, the region arrives at a presidential election with a law approved but no teeth.
Launches
-
Meta Business Agent on WhatsApp Business — A native AI agent for businesses on the dominant messaging channel in Latin America; available globally with a free tier. It enables customer service, scheduling and sales processing directly in the chat.
-
Microsoft MAI-Code-1-Flash in VS Code — A 5-billion-parameter model specialized in code; initially available to approximately 10% of GitHub Copilot users. To activate it in VS Code, just select the ‘Auto’ option in the model picker.
Threads we’re following
Since late May we have been following the entry of artificial intelligence into the Colombian presidential campaign: first came the use of avatars and synthetic voices in first-round ads, then the rushed approval of Law 2502 against electoral deepfakes. Today’s chapter closes that cycle with a legislative irony: the law exists, but its penalties start in July, when the runoff will already be over. Colombia thus becomes the first documented case in the region where an anti-deepfake law arrives too late for the election it was meant to protect.
If a Latin American regulator had to choose tomorrow between the European model (mandatory sovereignty, four cloud tiers, €420 billion in its own infrastructure) and the US model (30-day voluntarism, no prior controls), which of their country’s government data would fall into each category, and who would be making that decision?
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.