Halvar Flake says model outputs should not get copyright protection labs deny to books

max_paperclips · x · 2026-07-24

Halvar Flake argues that a human author cannot publish a book and then forbid labs from training on it under a fair-use theory, yet some labs appear to want the opposite protection for their own model outputs.

He frames the issue as a consistency problem: if training on copyrighted material is fair use, then banning training on model outputs would require a much broader rethink of fair-use assumptions.

Original post →

More from AGI Musings

AGI Musings channel →