On June 12, the US Department of Commerce did something unprecedented: it used an export control to order Anthropic to cut off access to two of its most powerful artificial intelligence models, Claude Fable 5 and Mythos 5. Since the company cannot distinguish in real time between US and foreign users, it ended up switching off both models for the entire planet. The directive and the company’s response mark the first time in history that a government has pressed the off switch on a frontier AI model from a private lab.
The stated reason, according to the order signed by the Commerce Secretary, is that an unidentified company reported having achieved a “jailbreak” of Mythos —that is, having circumvented its safety restrictions to activate capabilities the model should not offer. Anthropic describes the situation as a “likely misunderstanding”: it maintains that the jailbreak would be narrow (it would only be triggered by asking the model to read code and fix its vulnerabilities) and not a universal flaw. The rest of the Claude models are not affected.
For Latin America, the episode is a concrete lesson in dependence. The region did not take part in the agreements surrounding these models, has no seat at the G7 that begins on June 15 and was not consulted before the cutoff. Overnight, it lost access to Anthropic’s most capable tools without prior notice. Europe completes the paradox: this same week it presented a 420 billion euro package to ensure that “no one has a kill switch” over its technology infrastructure. The United States has just shown that its own works.
Also today
- The G7 brings together the CEOs of the big AI labs for the first time — The summit in Évian (June 15 to 17) seats the leaders of OpenAI, Google DeepMind and Anthropic at the table, just three days after the first shutdown of frontier AI. The United States is resisting multilateral agreements, and a watered-down final communiqué is expected, with no voice from the region.
- ChatGPT surpasses 1 billion monthly active users — It is the first app that is not a social network to cross that threshold. The milestone comes with conversational advertising already active in Brazil and Mexico, which profiles users based on their past conversations.
- A federal judge suspends Brazil’s fine against Meta over WhatsApp data — CADE, Brazil’s competition authority, will have to go to mandatory conciliation. The case will determine whether the country’s data law allows WhatsApp conversations to be used to train AI, a question that affects more than 400 million users in the region.
- SpaceX debuts on Nasdaq with a 19% gain — It closed at $160.95 per share and a valuation of $1.77 trillion in the largest stock market debut in history. Its compute contract with Anthropic ($45 billion over three years) now becomes mandatory disclosure information, and turns the compute of frontier labs into an asset accessible through the stock market for Latin American investors.
In the region
The directive on Fable 5 and Mythos 5 is the most concrete AI governance event of the week for Latin America: it shows that access to frontier models can be cut off from Washington in a matter of hours and without consulting the affected countries. The calendar is speeding up on several fronts. In Brazil, the data protection authority’s regulatory sandbox closes its consultations on June 15, just as the CADE-Meta case enters negotiation. In Colombia, a few days before the June 21 presidential runoff, five documented deepfakes are already circulating —the latest, a fake image of the Colombian national team spread during the World Cup’s opening match— while the law that should stop them will not be effectively enforced until July.
Threads we’re following
We had been following the arrival of Fable 5, the frontier model that Anthropic presented a few days ago together with its autonomous agents and its safety numbers. Today’s chapter gives it an unexpected twist: the same model that was being discussed for its capabilities has just become inaccessible not because of a technical decision by the company, but because of an order from the US government. The conversation about what these systems can do now intersects with another, more uncomfortable one for the region: who decides whether we can use them.
If access to frontier AI can be switched on and off from an office in Washington, what real room does Latin America have left to avoid depending on decisions made without 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.