Anthropic filed its confidential S-1 with the SEC on June 1, the first formal step toward an IPO planned for October 2026, at a valuation close to $965 billion. The same day, Microsoft held its Build 2026 conference in San Francisco and announced that Windows will become the first massive-scale platform to natively run artificial intelligence agents (programs that act autonomously to complete tasks).
The most striking announcement at Build was Project Polaris, Microsoft’s own model, which will replace GPT-4 in GitHub Copilot starting in August. For Latin America, where Microsoft dominates more than 80% of government and business computers, this means that the most widespread AI-assisted programming tool in the corporate sector will stop depending on OpenAI. In parallel, the Windows Agent Framework (released as open source) turns the region’s more than 200 million Windows PCs into potential execution nodes for AI agents with no need for additional cloud infrastructure. Satya Nadella summed it up on stage: “agents are not a feature.”
Anthropic’s S-1 also has a regional dimension: when the document is made public (estimated for between July and August) it will be the first time investors can see the full numbers of a frontier AI company. Goldman Sachs and Morgan Stanley are listed as underwriters of the process; the price and number of shares have not yet been determined.
Also today
-
OpenAI launches GPT-5.5, Codex and Managed Agents on Amazon Bedrock — OpenAI’s frontier models are now available with AWS’s regulatory compliance and billing, which makes it easier to adopt them in regulated sectors in the region such as banking, health and government.
-
Meta tests AI subscriptions ($7.99/month) in Guatemala and Bolivia — The first two Latin American countries where Meta is experimenting with a paid model to access AI features on Instagram, Facebook and WhatsApp. Meta Conversations takes place tomorrow, where AI news for WhatsApp could be announced.
-
Colombia heads to the runoff with an anti-deepfake law in force but no AI penalties — Law 2502 defines deepfakes as an offense, but the specific penalties for the use of AI do not take effect until July, that is, a month after the June 21 presidential runoff.
-
Colorado dismantles its AI law under federal pressure and tech lobbying — The anti-discrimination and impact assessment requirements were eliminated; what was the most ambitious state AI law in the United States was reduced to simple disclosure obligations, with implementation postponed to January 2027.
In the region
Colombia heads to the June 21 presidential runoff with anti-deepfake Law 2502 in force, but without the penalties for the use of AI being active: they take effect in July, when the most sensitive electoral period will already be over. It is the first real test in Latin America of an electoral AI rule with an approved law but no effective enforcement at the moment it matters most. What happens there will be a precedent for the whole region. In parallel, the dismantling of Colorado’s law under pressure from tech lobbying confirms a pattern Latin American regulators already know: the benchmark regulatory model will remain the European EU AI Act, not the US one. Also still pending is President Lula’s signature on Brazil’s PL 2338 (approved by the Chamber on May 27) and whether the copyright chapter, which requires compensation for the creators of content used to train models, survived into the final text.
Launches
-
Windows Agent Framework (Microsoft) — An open-source framework for running AI agents natively on Windows, with built-in sandboxing (an isolated environment to limit the agent’s access to the system) and governance. Already available on GitHub for developers; the local AI runtime arrives via Windows Update on June 9.
-
OpenAI on AWS Bedrock (GPT-5.5 + Codex + Managed Agents) — OpenAI’s frontier models available with AWS’s regulatory compliance and billing. Of interest to enterprise teams that already operate on AWS and want to compare options directly against other offerings on the market.
Threads we’re following
Brazil has been debating its AI bill (PL 2338) for months; it includes an article requiring compensation for the creators of content used to train models. The Senate approved it with that chapter included; on May 27, the Chamber of Deputies gave its approval. What is still unclear is whether the copyright text survived the legislative process or was amended, and whether the bill goes directly to President Lula or returns to the Senate for another round. This week is key to finding out.
Anthropic’s path to the stock market is another background thread: the company has raised record rounds over the past two years and its confidential S-1 this week marks the start of the final stretch. Once the document is made public (in 60 to 90 days) the markets will be able to see for the first time the full numbers of one of the fastest-growing frontier AI companies.
With the open-source Windows Agent Framework, OpenAI on AWS Bedrock and other frameworks available simultaneously this week, should Latin America’s technological sovereignty strategy be built on the “open” infrastructure of big tech, or is that openness simply another form of dependence that is harder to see?
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.