DAILY INTELLIGENCE BRIEF

The Vault — AI Edition

Sunday, August 30, 2026
16 stories
16,013 words
77 min read

Business

5
01

OpenAI Cuts Off Cursor After SpaceX Acquisition, Ending Access November 12

OpenAI has told Elon Musk it is taking its models back. On August 28, 2026, the company notified SpaceX that Cursor access ends.

5 minRead →
02

Anthropic Doubles Down on Cursor With Higher Claude Limits as OpenAI Walks Away

Anthropic needed about ninety minutes to answer the biggest question in developer tooling this week, and it answered with compute.

5 minRead →
03

Inside Project OT: How Meta Plan to Cut Teams 60% and Replace Them With Agents Collapsed

Project OT had a code name, a retreat, a playbook, and a date on the calendar. What it did not have was technology that worked.

5 minRead →
04

AWS Opens 900+ Cloud and AI Courses to Students in $500M Skills Push

AWS is giving away a $449-a-year training subscription to anyone who can prove they are enrolled at a university.

4 minRead →
05

Europe Grids Hit Their Limit as Speculative Data Centre Projects Clog the Queue

Britain energy regulator published a figure that reframes the entire debate about AI infrastructure in Europe.

5 minRead →

Llms Genai

3
06

MiniMax Pushes M3 to General Availability, Adding a 428B MoE to the Open-Weight Flood

MiniMax spent the summer doing what Chinese AI labs now treat as routine: ship a frontier-class model behind an API, then open it.

4 minRead →
07

ChatGPT Temporary Chats Get Memory, Plugins and a Save Button

For two years ChatGPT Temporary Chat behaved like a burner phone: nothing came in, nothing went out, nothing stayed.

5 minRead →
08

The Producers Turned Detectives: Inside the Hunt for AI-Generated Songs on Streaming

Max Harris listens for the hiss. The EDM producer has spent months turning his feed into an informal forensics desk.

5 minRead →

Research

3
09

DeepMind Cyclone Model Buys Forecasters a Full Extra Day, Nature Study Finds

Google DeepMind published in Nature the peer-reviewed case for a model that had already been in operational use.

5 minRead →
10

LLMjacking Evolves: Attackers Are Now Using Stolen AI Compute to Build Offensive Agents

A residential broadband connection began sending requests to an AI model server that should never have been reachable.

5 minRead →
11

Multi-Agent Science Moves From Demo to Method, and the Bottleneck Moves to Verification

The claim that AI can do science stopped being a press release in 2026 and became a citation.

5 minRead →

Policy

5
12

Sony Music and Warner Chappell Sue Anthropic, Alleging Piracy at the Root of Claude Training Set

The music industry has spent nearly three years arguing that training an AI model on copyrighted songs is itself infringement. This suit changes the question.

4 minRead →
13

The Fed Starts Treating AI as a Factor of Production

On his 100th day in the job, Kevin Warsh spent the opening of his Jackson Hole keynote on artificial intelligence.

5 minRead →
14

Poland Asks Brussels to Fine Meta 250M Euro Over Scam Ads; Meta Answers With an AI Detector

Poland cybersecurity agency ran a test on Meta. CERT Polska flagged 122 advertisements it had classified as fraudulent.

5 minRead →
15

Siemens CEO to Brussels: Your Rulemaking Takes Two Years, AI Ships Eight Generations

Roland Busch did not measure Europe regulatory lag in months. He measured it in model generations.

5 minRead →
16

xAI v. Colorado: The First Real Test of Who Governs American AI

xAI is asking a federal judge in Denver to answer a question no court has squarely resolved: who gets to govern AI.

5 minRead →