What this week looked like a technical dispute turned out to be, above all, personal. On Monday, June 15, senior Anthropic executives met in Washington with Trump administration officials to try to reverse the order that cut off global access to its most powerful models, Claude Fable 5 and Mythos 5. That same day, Axios revealed that the origin of the cutoff was not a safety protocol but a falling-out between people. As context: a few days ago the US government used an export control to “switch off” those two models around the world, Latin America included, after a way to get around their safety barriers (a “jailbreak”) was detected.
According to Axios’s reconstruction, when the government warned Anthropic about the problem before acting, CEO Dario Amodei declined to describe it as a serious risk and refused to fix it or withdraw the model voluntarily. David Sacks —Trump’s artificial intelligence adviser— went so far as to describe the company’s stance as “blackmail” of the government, as reported by Tom’s Hardware. Anthropic responds that it worked with federal agencies on pre-launch testing and that the jailbreak is not universal. The meeting, reported by CNBC, resolved nothing: the ban remains in force with no timeline for resolution, while a Semafor report claims that a Chinese group may already have accessed the compromised version.
For Latin America the reading is uncomfortable and direct. The region’s access to the most capable AI models on the planet does not depend on a multilateral mechanism, a negotiation between states or a process with a right of appeal, but on whether the head of a lab and a US government adviser manage to get along. Whoever was not at the table where the decision was made also has no way to influence when it is lifted.
Also today
- OpenAI could capture the week with GPT-5.6 — the prediction market Polymarket gives an 83% probability of a launch before June 30; with Fable 5 blocked and Google’s Gemini 3.5 Pro without general availability, OpenAI would have the frontier field to itself.
- Second lawsuit against Anthropic in two days — a class action filed on June 14 in California alleges that the Claude Max plans ($100 and $200 a month) offer less usage than promised; the class would include subscribers in the region.
- SpaceX closes its second week on the stock market up 31% — the stock (SPCX) is the first stock market asset directly tied to the compute of a frontier lab: its cloud contract with Anthropic is now public information, and any turn in the ban moves it.
In the region
Five days before Colombia’s presidential runoff, several oversight institutions declared almost at the same time that AI-generated disinformation has exceeded their capacity to respond. The Electoral Observation Mission published three formal warnings about the digital campaign, the polling firm Invamer had to debunk a fake poll fabricated with AI, and the only rule available, Law 2502, will not have enforceable sanctions until July: it exists, but it is inoperative precisely during the country’s most digitalized campaign. Meanwhile, Brazil is moving in the opposite direction and is building up the most active regulatory framework in the region: the telecommunications agency ANATEL opened a public consultation on AI and cybersecurity with a deadline of June 25, which adds to the AI Legal Framework awaiting presidential signature. And in Chile, Washington’s shutdown turned a theoretical debate into a proven hypothesis: a column in Estado Diario uses the case to discuss digital sovereignty, noting that the few regional equivalents —the LatamGPT project of the Chilean center CENIA and the Mexican Coatlicue plan— remain the exception.
What institution —regional, multilateral or yet to be invented— would today have the authority and technical capacity to respond both to the shutdown of a model decided between two people in Washington and to an election in which the state has already admitted that disinformation overwhelmed it?
Correction (September 30, 2026). The original version described David Sacks as “CEO of xAI, a direct competitor of Anthropic” and said he “also runs the competition”; in fact, Sacks does not run xAI: he is co-chair of the President’s Council of Advisors on Science and Technology (PCAST) and was the White House AI czar, according to Tom’s Hardware.
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.