AI weekly: 2–8 November 2026
Every article published from 2–8 November 2026, newest first, with a short summary of each.
This week's articles
Kimi K3.1 Pro and ERNIE 5.5 Flash Base: sizing giant open MoE models
Moonshot and Baidu published very large mixture-of-experts weights this week. Here is how MoE works, what the licences allow, and how much memory hosting needs.
Governing AI agent access with scoped credentials and undo buttons
A reported Palo Alto deal, Goldman's tool limits and a deleted syllabus show why AI agent credentials and permissions are becoming their own security category.
GPT-6 Cyber, Tencent Hy4 and DeepMind's RSA work on AI bug hunting
OpenAI, Tencent and Google DeepMind all shipped security-focused AI work this week. What happened, how AI bug hunting works, and what defenders should do now.
Always-on AI agents move from chat sessions to background teammates
Perplexity, Meta, DeepSeek and Airbnb moved agents towards persistent background work this week. How long-running agents are built, governed and budgeted.
AI's bottleneck moves to chip packaging, optics and idle GPU capacity
This week's AI infrastructure news in plain words: why chip packaging and optical links now limit supply, and how GPU partitioning helps you use what you have.
Rust for AI: fast, safe inference, data tooling and where Python still wins
Where Rust fits in AI systems: inference with Candle and ort, data tooling with Polars and Arrow, PyO3 extensions, WASM, and why research stays in Python.
Go for AI: where it fits, the libraries, and the patterns that work
Where Go fits in AI systems, which Go libraries and SDKs to use for models, agents, MCP and vectors, and three tested patterns for serving, tools and batching.
In-context imputation for tabular data: filling missing values with a foundation model
How in-context imputation fills missing values in tables, how it compares with MICE, where it loses, and how to evaluate any imputer honestly.
An In-Context Imputation Head for Tabular Data: Chained, Validation-Gated, and Evaluated on Three Imputation Boards
Paper draft: a chained in-context imputer with a validation gate, results on three imputation boards, its losses, and seven evaluation defects.