irl.works found that Indian digital public infrastructure systems fall short of the Indian government’s own stated claims and that none of the six systems studied had an active forum for public participation.
The IMF reports that extreme cyber-incident losses could trigger funding strains, raise solvency concerns, and disrupt broader markets, and that Mythos could find and exploit vulnerabilities in every major operating system and web browser even when used by non-experts.
OpenAI announced it is slowing the scaling of some advanced training runs and took a two-week pause in reinforcement learning on its latest models intended for release.
Kuhle Daweti writes that the US and China host about 86% of the world’s data centre capacity, the US accounts for about three quarters of the high-performance computing used to train advanced AI, and that until South Africa, the greater African continent and Brics countries, China aside, own more of the infrastructure AI runs on, their sovereignty will be rented.
The authors of “LAAF: A Layered Accountability Architecture Framework for LLM Applications” report that a systematic review of 122 primary studies and 12 regulatory documents identified four persistent gaps and five structural tensions, and consolidated a classification device into LAAF, an accountability architecture aligned with the OWASP LLM Top 10 (2025), as a synthesis of surveyed evidence.
‘Which Rules Matter Now? Policy-Centroid Routing Before an Intelligent System Acts’ reports that policy-centroid routing creates a layer before adjudication, its output is a review agenda, and it includes a synthetic worked example with no empirical efficacy result.
Today’s lead: Brazil and Portugal account for about 69% of Portuguese dataset records.