Lean and Mathlib certify the Claude-assisted Riemann proof, OpenRouter off-boards three inference providers after canary tests exposed data leaks, and Upload2Shell turns Ghostscript uploads into code execution.
Daily News – 21 November 2026: Claude-assisted Riemann proof claim
An Anthropic researcher claims a Claude-assisted Lean proof of the Riemann Hypothesis, attacker-run Cairn AI agents found and exploited a zero-day on their own, and Thinking Machines released the 20 GB open Inkling 2 Mini.
Daily News – 20 November 2026: Anthropic goes public as agent guardrails come under scrutiny
Anthropic begins trading on Nasdaq, Check Point warns that weak AI guardrails fail against strong models, and Microsoft ships the open-weights Aion Instruct inside Edge.
20 Watts vs 10 Kilowatts: How Far Is AI from the Human Brain?
Your brain runs on about 20 watts. The server that runs today’s best open AI models needs hundreds of times more. A comparison of connections, energy, speed and learning, from Hinton’s warnings to brain-inspired chips.
Six Outages, One Week, Very Few Answers: What the Companies Have Actually Explained
Between 12 and 17 November 2026, Microsoft, AWS, the FAA, Chase, United, Apple and the NSA all had major outages. Every official says it was not an attack. This post does not argue that it was. It asks a narrower question: how much has each organization actually explained?
Daily News – 19 November 2026: Gemini 4 Neon and a critical NGINX HTTP/2 flaw
Google ships Gemini 4 Neon at $3/$15 per million tokens, NGINX gets two critical HTTP/2 RCE fixes, and Microsoft’s Agent Framework package on PyPI was hijacked to steal credentials.
Daily News – 18 November 2026: OpenAI ships GPT-6.2 Astra and Sol
OpenAI launches GPT-6.2 Astra and Sol, the Java4Shell OpenJDK flaw found with Codex needs patching, and Microsoft ships MAI Code 2.
