Aug 30, 2026
Today in brief.
Sony and Warner sued Anthropic over tens of thousands of songs
The two biggest music publishers filed a multi-billion-dollar copyright suit against Anthropic and its founders, alleging Claude was trained on torrented and scraped lyrics at a scale no prior AI music case has claimed.
Claude Code limits are going up 25%, not down, and HN got it backwards
Anthropic is permanently raising Claude Code weekly limits by 25% from Sept 14 - the HN thread that hit the front page said the opposite, so this is a check-the-source story.
Altman says OpenAI will call something AGI by December, and it's Astra
In the TIME profile, Altman says OpenAI expects an internal system it would call AGI before year-end, Mark Chen puts them at 80% of the way, and the model behind the claim is Astra - the same one from the Hugging Face and Cursor stories.
A 1,250-worker survey found 3% lost a job to AI and 9% got promoted by it
Four years after ChatGPT, a July-August survey of 1,250 employed US workers puts AI-caused job loss at 3%, AI-created jobs at 6%, and AI-driven promotions at 9% - the gains slightly outnumber the losses.
Open source is still fighting about AI, and Debian just voted
A widely-shared essay maps why maintainers keep banning AI contributions - and on the same day Debian's general resolution landed on "responsible use", with the ban-and-expel options losing to None of the Above.
Qwen 3.8 27B runs at 50 tok/s with 100k context on a 16GB GPU
The local-model corner's win this week is a 27B Qwen at 50 tokens per second with a 100k window on a consumer 16GB card, via the beellama.cpp fork - plus a curated list of the llama.cpp PRs that make CPU and hybrid inference faster.
Warp published the pattern for self-improving agents on Claude
Warp wrote up the development loop it uses to make its agents improve themselves on Claude - pitched as a simple pattern anyone can copy, and it made HN's front page.
No stories in this category for this edition.