OpenAI launched a science blog with a heavyweight opening result, opencode shipped a meaningful change to how subagents run, and Sebastian Raschka published a deep dive on positional embeddings. On the security side (7-day incident window), Cursor explained how its Claude Code extension compromise happened, and Google patched an eighth actively-exploited Chrome zero-day.
From the labs
- A step toward defining the Standard Model on a lattice — OpenAI (+ science blog launch)
- opencode v2.3.0 — opencode (background subagents by default)
Worth your time
- Understanding Positional Embeddings in Modern LLMs — Sebastian Raschka
Beyond AI
- An update on our recent security incident — Cursor (+ The Hacker News)
- Stable Channel Update for Desktop — Google Chrome (XML zero-day, exploited in the wild)