Mistral Large 4, an open-weight model trained in Europe, owned the day and framed the bigger question: who controls the AI other people run. The same thread runs through OpenAI publishing machine-made mathematics, a mega media merger clearing its last legal hurdle, and a privacy fiasco at Meta.
Canada tests a new door into Europe while the tools that shape trust get a closer look — from cameras on poles and photos on phones to browsers, app stores, databases, and the way we train AI itself.
A four-page math statement and a 10,000-agent proof put verifiability in the spotlight, while TVs, CDNs, wikis, and office software each test who holds control when capability gets faster, cheaper, and more concentrated.
When a single browser engine, bulk-made buying guides, and bulk-harvested IDs shape what we click and trust, cheaper and faster is not automatically more trustworthy.
Meta put a 30-billion-parameter agent on a single consumer GPU, Docker boxed coding agents into disposable VMs, and a meeting recorder leaked 180,000 sessions. Illinois also pushed age checks into the operating system.