The picture of global artificial intelligence governance shifted a little in the region’s favor today: it was confirmed that Costa Rica co-facilitated, together with Spain, the preparatory process for the first UN Global Dialogue on AI Governance, which opened this week in Geneva. It is a second Latin American institutional leadership role, adding to that of El Salvador, whose ambassador Egriselda López co-chairs the dialogue together with Estonia. Yesterday we reported that the region arrived with “one seat”; today there are, properly speaking, two.
The caveat comes on the same day, and it is a harsh one. The Independent International Scientific Panel advising the process, co-chaired by Yoshua Bengio and Nobel Peace Prize laureate Maria Ressa, put a figure on the table: 118 countries, most of them from the Global South, are effectively left out of the substantive AI governance discussions. Two leadership seats do not cancel out that underlying asymmetry, and no Latin American government confirmed a ministerial delegation of its own, unlike India. At the opening, Secretary-General António Guterres called for banning lethal autonomous weapons and warned against letting AI “vibe-code” humanity’s future. Meanwhile, Sam Altman pressed his idea of creating an AI safety body modeled on the International Atomic Energy Agency, a proposal at odds with the multilateral process that the UN, UNESCO and the International Telecommunication Union are pushing this very week.
Also today
- Anthropic describes an internal “workspace” in Claude that evokes theories of consciousness — A technique dubbed “Jacobian Lens” detects an activation pattern that anticipates words the model has not yet written; the company itself urges caution about sensationalist readings.
- Dario Amodei qualifies his “Machines of Loving Grace” thesis — Anthropic’s CEO admits that the pace he promised, “a decade of science per year,” has not yet materialized in biotechnology.
- Anthropic signs a 20-year lease worth about $19 billion with TeraWulf — About 401 MW at a data campus in Kentucky, with first capacity expected in 2027: another sign that the AI race is increasingly played out in electricity and bricks.
- Memory chip workers are South Korea’s most sought-after bachelors — The salary boom in HBM memory for AI, with bonuses of up to $476,000, is reshaping even the dating market.
- RAND: winning the AI “pentathlon” takes endurance, not a sprint — An essay from the RAND Corporation argues that AI leadership is decided in a long-distance race across compute, data, talent, energy and governance, not in a single technological milestone.
In the region
The operational side of the issue showed up in three pieces today. First, the driving force behind Latam-GPT explained why AI models make more mistakes when they talk about Latin America: between 20% and 40% more “hallucinations” (made-up answers that sound plausible) because of the scarcity of regional data in training, a concrete argument for data sovereignty. Second, AWS warned that 90% of AI projects in the region do not get past the proof of concept, and attributes the failure to organizational execution more than to technology. And third, a sign that the region is already buying and deploying: Spain’s Substrate AI won a contract with the Buenos Aires City Government for its platform to be the foundation of BAX, the Buenos Aires city government’s AI assistant.
If models make more mistakes when they talk about us, and the region is only just emerging with two leadership seats but no ministerial delegation of its own, is it up to Latin America to build its own table, or will it keep negotiating at other people’s?
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.