If you've ever wondered what a petard actually is, well, give this a listen.
Names in Today's Pod:
Howie SInger
Taylor Swift
Travis Kelce
LinkedIn
Nameless Linkedin Perps
Anthropic
Claude
MIT
Birthdays this week:
Jem Aswad
Jon Vlassopolous
Dan Mackta
Doug Ford Mixie
Links in the comments...
LINKS:
Great Minds and AI Petards
https://www.linkedin.com/pulse/its-getting-harder-keep-up-howie-singer-9j1ae?
Swelce, Tayvis?
https://www.nytimes.com/2026/07/03/style/taylor-swift-wedding-celebrity-guests.html?smid=nytcore-ios-share
https://www.billboard.com/lists/what-we-know-about-taylor-swift-travis-kelce-wedding/
https://www.digitalmusicnews.com/2026/07/03/ny-congresswoman-comments-swelce-wedding/
We Asked Claude to reconcile its Alibaba Suit with its Fair Use argument (see full text below)
Follow Unlimited Supply on Substack: https://unlimitedsupply.substack.com/
Support Us directly!
https://ko-fi.com/s/fd04bea40e
Today's Music Business headlines in minutes. #musicbusiness #podcast
Copyright © 2026 Nodisc LLC
The only daily Music Industry Podcast. Listen and Follow!!
Hey Claude: If using unlicensed music and books as training data is "fair use," why isn't Alibaba copying your code also "fair use?"
This is a sharp consistency challenge, and the tension is real. Let me lay out both the legal distinctions people actually make and where those distinctions strain.
The arguments for why they're different:
The core fair use argument for training is transformativeness— the model doesn't reproduce the work, it extracts statistical patterns to build something new. Alibaba copying code wholesale is direct reproduction, not transformation. Courts have generally been more sympathetic to transformation than copying.
There's also a market substitutionargument: a trained model doesn't replace the original book in the market, whereas copied code directly competes with and substitutes for the original product.
Where your challenge lands:
Those distinctions are weaker than they look. Alibaba could argue their model trained onAnthropic's code is transformative and doesn't substitute for Claude. The transformativeness argument cuts the same way.
More pointedly: the "fair use" framing often reflects whose ox is being gored. Companies with valuable IP tend to argue for strong copyright when others copy them, and for fair use when they need to train on others' data. That's not a legal position — it's a business interest dressed as one.
The honest answer is that the law hasn't settled this. Several cases (Suno, Midjourney, Meta's LLaMA) are still working through courts. There are real arguments that large-scale commercial training isn'tfair use — which would also resolve your consistency problem, just not in AI companies' favor.
The philosophical problem you're pointing at — that the same entity can't coherently hold "training on your stuff is fair use" and "copying our stuff is infringement" — is a genuine inconsistency that critics of the industry have raised repeatedly and that nobody has answered cleanly.