What is new today changes the meaning of the story: a senior White House official confirmed that the cutoff of access to the Claude Fable 5 and Mythos 5 models is specific to Anthropic and will not be extended to other labs, and that it was Amazon that alerted the administration to the alleged “jailbreak” of Mythos, as revealed by The Information and reported by Fortune. The detail is uncomfortable, because that same Amazon hosts Fable 5 on its Bedrock platform and lost the service along with the rest of the world. Let us recall the context: since June 12, an order from the US Department of Commerce has forced Anthropic to switch off those two models for the entire planet, in the first case of a government using an export control as an off switch on a frontier AI model.
That the ban is aimed at a single company, rather than an even-handed rule for the whole industry, moves the question from the terrain of national security to that of corporate geopolitics. No one has publicly explained what incentives Amazon —Anthropic’s compute provider— had in warning about a vulnerability in a model it hosts on its own cloud. Anthropic, for its part, maintains that the standard applied, if generalized, “would halt all frontier model deployments across the entire industry.” The ban remains in place with no announced timeline for resolution, and the international reaction has been strong: executives and policy analysts accuse Washington of turning access to AI into a tool of geopolitical pressure. For Latin America the reading is direct: the region lost access not because of a universal policy, but as collateral damage from a targeted political decision on which it was not consulted.
Also today
- The Évian G7 opens with a rift between the United States and the European Union over AI sovereignty — It is the first G7 with the three chief executives of the big frontier labs together (Sam Altman, Dario Amodei and Demis Hassabis), invited by the French president. Washington is resisting binding agreements; Brussels arrives with a 420 billion euro plan for technological sovereignty. A watered-down final communiqué is expected, and no Latin American government is at the table.
- Colombia reaches the runoff with a fake audio of the Registrar still circulating — The final Guarumo and Ecoanalítica poll puts Abelardo de la Espriella at 52.6% against 45% for Iván Cepeda, one week before the June 21 runoff. The Registrar asked people not to spread false information while a deepfake audio of his voice remains active on social media, and the electoral law against these practices does not take effect until July.
In the region
The Évian G7 opens as the first multilateral forum after the ban, and it exposes a gap: the fact that the cutoff is aimed at a single company and was triggered by a private provider adds a layer of corporate geopolitics that the G7 has no mechanism to resolve. Is it a precedent for export policy, an episode in the dispute between Anthropic and the Pentagon, or a miscalculation by the White House? Meanwhile, in the region the clocks are running on their own: in Colombia, the June 21 runoff is the next test of AI in an election, with the Registrar’s deepfake still unpunished; in Brazil, the data protection agency closed the consultation on its regulatory sandbox yesterday. The immediate milestones are June 17, with the G7’s final communiqué, and June 21, with the Colombian runoff.
Threads we’re following
This adds to the story we have been following since June 12: the first shutdown ordered by a government on a private frontier AI model. Yesterday the question was whether the cutoff would set a standard for the whole industry; today the official answer is no, that it targets only Anthropic, and that the warning came from Amazon. The story stopped being about a technical vulnerability and became about who decides, and with what incentives, the world’s access to a key technology.
If access to frontier AI can be switched off by someone else’s foreign policy decision, what tools do regulators in Brazil, Colombia, Chile or Mexico have to keep it from depending on third parties?
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.