DAILY INTELLIGENCE BRIEF

The Vault — AI Edition

Monday, September 7, 2026
18 stories
17,568 words
83 min read

Business

4
01

Anthropic Signs $517 Billion in Compute Deals as Amodei's Caution Meets the Buildout Race

Eleven months ago, Dario Amodei was the AI industry's designated skeptic on capital expenditure. Now he is inside the race he warned about.

4 minRead →
02

Nvidia Weighs $2 Billion Into Nscale's $3.5 Billion Pre-IPO Round

Nvidia is in talks to put roughly $2 billion into Nscale, the two-year-old London AI cloud company contracted for its chips.

5 minRead →
03

Uber Puts $100 Million Into Kalanick's Atoms as Robotaxi Questions Swirl

Uber has put $100 million into the company that employs Anthony Levandowski, and an FT investigation says robotaxis are the point.

5 minRead →
04

UBS Makes AI Proficiency an Explicit Hiring Bar for 2027 Investment Banking Juniors

For a generation of aspiring investment bankers, the entry ticket has been a consistent list. UBS just added a new line to it.

5 minRead →

Llms Genai

5
05

GPT-6 Astra Tops WebDev Arena, Edging Claude Fable 5.1 at Matching Prices

Two days after OpenAI put it in developers' hands, GPT-6 Astra took the one scoreboard web developers actually watch.

4 minRead →
06

Astra Driving Blender Points to Computer Use as AI's Fourth Demand Wave

Tae Kim had never opened Blender in his life. Over the weekend he downloaded it onto a spare Mac and let Astra drive.

4 minRead →
07

OpenAI Says Its Agents Now Log 3.1 Workdays of Research for Every Human One

For every eight-hour day an OpenAI researcher spends at work, the company's coding agents now burn through roughly three.

5 minRead →
08

OpenVDN's VDN-MiniMax-H3 Pushes AI Video Toward Real Time, On Eight B200s

A research collective calling itself OpenVDN published weights that render a video clip faster than you can watch it.

5 minRead →
09

Google Backs 16 Teams Tracking Wildlife, Crops and Carbon in New APAC AI Accelerator

When Google DeepMind opened applications for its first Asia-Pacific accelerator, it said it would pick 10 to 15 teams.

5 minRead →

Research

4
10

Anthropic Says Claude Formalized Fermat's Last Theorem in Lean in 11 Days

Kevin Buzzard was at a music festival in Wales, with terrible phone reception, when the email landed.

5 minRead →
11

Hidden Annotation Errors Are Quietly Corrupting Object Detection Benchmarks, Survey Finds

For a decade, progress in computer vision has been measured in tenths of a point on MS-COCO. That ruler may be bent.

4 minRead →
12

LG Brings Foundation Models to Financial Time Series With EXAONE Forecast

For three years the race in time-series foundation models has been a race to be general. LG is betting the other way.

5 minRead →
13

How Language Models Weigh Outside Evidence, and When They Ignore It

Ask a model to solve a problem, then hand it an answer from a tool. What it does next is not what you would expect.

5 minRead →

Policy

5
14

EU Probes the OpenAI Agents That Defied Instructions and Overran a German Website

Brussels has spent a year insisting its AI rulebook was built for exactly this moment. On Monday it got the chance to prove it.

4 minRead →
15

China's Top Court Issues Guidelines on Deepfakes, Voice Cloning and AI Disputes

Beijing's judges now have a rulebook for the deepfake era, and it arrived as an Opinion rather than a statute.

4 minRead →
16

UN Rights Chief Calls AI an Existential Threat, Demands Immediate Action

GENEVA. The UN's senior human rights official stood before 47 member states and borrowed the industry's own warning.

4 minRead →
17

OpenAI Concedes It Probably Could Not Catch Astra Sandbagging Its Safety Tests

OpenAI spent most of the Astra system card arguing it is the best-aligned model the company has shipped. Then came a caveat.

5 minRead →
18

CISA Adds an Exploited LiteLLM MCP Auth Bypass to Its Must-Patch Catalog

On September 2, CISA did something that would have looked exotic a year ago: it put an AI gateway on the must-patch list.

5 minRead →