The US ban on the most advanced artificial intelligence models opened a crack for the first time, but only for a closed group. US Commerce Secretary Howard Lutnick sent a formal letter on June 26 authorizing Anthropic to give access to its Mythos 5 model to more than a hundred organizations classified as “trusted partners” (Fortune 500 companies and designated federal agencies), while the other blocked model, Fable 5, remains unavailable to the general public (aitoolly · CNN). It is the first partial resolution of a blockade that has now lasted fifteen days: early this month Washington banned two Anthropic frontier models over a safety risk, and since then the region has not been part of the conversation.
What matters is not only that access is coming back, but how it comes back. The restoration is “America first”: the more than one hundred authorized organizations are large US companies and federal agencies, and there is no Latin American organization in this first round. The region did not take part in the prior negotiations, was not consulted, and has no mechanism to influence who gets onto that list of “trusted partners.” It is also the first time the US government has explicitly put into practice a registry of organizations approved to use frontier models, the most capable AI systems on the market. If that mechanism becomes permanent, the question for Latin America is no longer when access returns but whether there is any institutional path onto the list. Today there is none.
The timing underscores the gap. The arrangement is formalized just one day after the region’s technology ministers adopted, in Santo Domingo, the Declaration of the UNESCO summit on AI ethics, their first attempt to set a common position. While the region produces its governance roadmap, Washington is rolling out a tiered system of access to the most capable models that, for now, does not include it. Access to OpenAI’s GPT-5.6 also remains under a preventive restriction: two of the most advanced systems in the world remain wholly or partly out of the region’s reach.
Also today
- Anthropic accuses Alibaba of the largest known “distillation” operation against its Claude model — In a letter to the Senate Banking Committee, the company describes 28.8 million exchanges through 25,000 fraudulent accounts to copy its model’s capabilities. Two senators announced a bipartisan amendment to sanction that kind of access; if it moves forward, it would be the first legal framework on copying AI capabilities, with possible effects for heavy users of these tools in the region. It is worth noting that Anthropic had proposed this very measure and has an IPO in the works.
- OpenAI is reportedly weighing postponing its IPO until 2027 — According to Bloomberg, growing regulatory complexity would push back the stock market debut; SoftBank shares fell on the news. It is a single source, without confirmation from the company.
- OpenAI retires its GPT-4.5 model and consolidates its lineup — Users move automatically to the next model while its most capable system is held for July under restriction. A TechCrunch analysis argues that the pattern “is no longer Anthropic versus OpenAI”: the US government is becoming the gatekeeper of access to frontier models across the whole industry.
- Colombia’s Telepatía AI raises $33 million in a round led by a16z — The company presents it as the largest digital health round in the region’s history. Founded in Colombia and active in more than 25 hospitals in Brazil, Colombia and Mexico, it says it serves 14 million patients (self-reported figures).
In the region
An investigation by The Intercept documented that the Pentagon, through its Special Operations Command South, runs a platform called “La Tilde”: a Spanish-language news site aimed at Latin American readers and produced with generative AI, with no identifiable authors and with the only notice of public funding buried in the site’s information page (The Intercept). Atlantic Council researchers describe it as “artificial intelligence from start to finish,” and the site reportedly plans versions for Ecuador, El Salvador, Guyana, Honduras, Jamaica, Panama and Peru. The contrast is hard to ignore: the same administration that blocked a model because of its danger to critical infrastructure uses generative AI at scale to produce messages aimed at the region. Organizations such as the EFF and Privacy International warn that the notice complies with the law but does not guarantee transparency, because the cost of producing geotargeted propaganda with these models is approaching zero.
If the next frontier of AI access is defined by lists of “trusted partners” decided in Washington, what would need to exist (prior consultation, an audited sandbox, formal representation) so that Latin America is not, once again, the last to find out?
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.