Daily AI RoundupsSunday, 30 August 2026

Sony Music and Warner sue Anthropic over alleged AI-training infringement as SpaceX moves toward turbine-blade production for data-center power. Sungho Park and coauthors report AutoSaddler improves agent benchmarks through harness optimization, while Zartbot analyzes OpenAI's Jalapeño inference architecture. The Information reports Cognition at about $900 million in annualized revenue with possible $800 million burn. Anthropic raises Claude Code limits, Debian approves responsible generative-AI use, and open-source releases span news framing, usage monitoring, dataset extraction and ThoughtDAG context graphs.

• AI copyright
• data-center power
• agent harness optimization
• AI chips
• AI developer tools
• open source
• model debugging

Stories in this roundup

1. 01
   Sony Music and Warner sue Anthropic, alleging a brazen campaign of intellectual property theft.TechCrunch
2. 02
   SpaceX lays groundwork for a turbine-blade factory to bypass data center power bottlenecks.The Information
3. 03
   Sungho Park and coauthors from Microsoft, POSTECH, KAIST and SUSTech report AutoSaddler raised agent benchmark scores by 9.0–10.0 points through automatic harness optimization.arXiv
4. 04
   Zartbot’s Hot Chips deep dive argues OpenAI’s Jalapeño targets low-latency inference with 64 local-memory slices, dedicated collective links and AI-assisted XLS chip design.Zartbot
5. 05
   The Information reports Cognition reached about $900 million in annualized revenue but could burn $800 million this year as Nvidia-server spending rises.The Information
6. 06
   Anthropic will permanently raise weekly Claude Code limits by 25% on Sept. 14, ending a 50% promotion.Claude Developers
7. 07
   Debian votes to allow responsible use of generative AI in project work.LWN
8. 08
   422August open-sources news-deframe to compare agency, claim coverage and wording across outlets reporting the same event.GitHub
9. 09
   Usage Notch adds a macOS edge overlay for Claude Code and Codex rate-limit windows using local CLI records.GitHub
10. 10
    Danilo Šegan tests seven frontier models on a native-PostScript pelican-on-a-bicycle prompt, with Gemini 3.7 Flash and Opus 5 correcting rendering errors through iteration.Danilo Šegan
11. 11
    Dataset Cleaner CLI 1.0.0 releases a pipeline for extracting question-and-answer datasets from documentation.PyPI
12. 12
    ThoughtDAG v0.3.33 updates its editable graph interface for making LLM context visible and user-controlled.GitHub