Court documents declassified on July 2 in the litigation between Anthropic, the lab behind the Claude assistant, and the US Department of Defense reveal a sequence that is hard to explain: one day after the Pentagon finalized the company’s designation as a “supply chain risk,” and before even notifying it, the Under Secretary of Defense who was negotiating with it described the talks as “very close.” Tech Times reports it based on the emails in the case file.
The judge in charge of the case found that exchange hard to reconcile with the government’s argument that Anthropic posed a “hostile and intolerable” risk, and sees signs of possible retaliation for the company’s refusal to give up full control over the military and surveillance uses of its model. Beyond who is right in court, the episode raises a question that goes beyond Washington: how discretionary (and potentially instrumental) a “national security” label can become when it is used as leverage in a commercial negotiation with an AI provider.
For the region, the precedent is concrete. Any Latin American government negotiating today with a frontier lab over the military, police or surveillance use clauses for the AI it buys for the state is watching, unintentionally, a dress rehearsal of the asymmetries at play: on one side, a buyer with the power to reclassify its provider overnight; on the other, a company trying to set limits on how its technology is used. It is worth following how it ends, because it sets the tone for contracts not yet signed.
Also today
- Micron breaks ground on a $9.3 billion expansion in Hiroshima — The new plant will make HBM memory for AI, with Japan subsidizing about $3.1 billion. A sign that the compute bottleneck is shifting from graphics cards to memory.
- India confirms a ministerial delegation for the UN Global Dialogue on AI Governance — The meeting is in Geneva this week; the contrast with the Latin American representation confirmed so far is striking.
- “From the mañaneras to AI training” — A Mexican analysis describes how the government’s official communications could seep into the data used to train language models, a form of contamination known as “LLM grooming.”
- Tesla sets a cap of $200 a week per employee on AI consumption — The measure starts on July 6 and joins AT&T and Amazon: large tech employers that are already containing spending on “agentic” AI after the shift to usage-based billing.
In the region
All the attention in governance this week is on the UN Global Dialogue on AI Governance (July 6 and 7 in Geneva), co-located with the International Telecommunication Union’s AI for Good summit. India has already confirmed a dedicated ministerial delegation, while Latin American representation remains limited to a Chilean academic spokesperson within the scientific panel, with no government from the region having a confirmed delegation of its own as of the close of this summary. In parallel, it was clarified that the OECD Employment Outlook 2026, expected on Tuesday the 7th, would not include, unlike the 2023 edition, a chapter specifically dedicated to AI, although it will address “technological shocks” in employment; it is worth checking the full table of contents on launch day before treating it as a reference report.
Launches
- Claude Science (beta) — A workspace with several AI agents aimed at scientific research (genomics, proteomics, cheminformatics), available in paid beta since July 1. It may interest universities in the region with biomedical research lines, although access through a paid plan is not designed for public budgets.
Threads we’re following
The Latin American vacuum in Geneva does not come out of nowhere. This same week we were following how a UN scientific panel put numbers on the concentration of the compute that trains the most capable systems: the vast majority is in just two countries, and the region barely appears on the map. The Geneva meeting is the political chapter of that same story: when India sends ministers and Latin America sends an academic with no government rank, the distance is no longer only technical and becomes a matter of the negotiating table.
If the power to negotiate the rules of AI is distributed according to who has compute, a market and a delegation of their own, what does a country that has none of the three bring to the table?
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.