NewsBite analysis · AI, Regulation
The Decoder
US Copyright Office says fair use does not cover AI trained on "vast troves of copyrighted works"
In a recent report, the US Copyright Office concluded that the fair use doctrine does not extend to AI systems trained on large datasets of copyrighted works. The agency dismissed claims likening AI training to human learning and highlighted that generative models often reclaim copyrighted expressions rather than simply recognizing patterns. While the full report allows for limited exceptions based on transformative use, it notably emphasizes the significance of the originality and legality of the training data. Illegal sourcing notably undermines fair use claims, suggesting ongoing scrutiny within the AI sector regarding copyright compliance.
MonitorConfidence: HighShort term horizonRegulatory risk: High
Published 22 Jul 2026, 20:10 UTC1 min read
Visit original article ↗Lead image — The Decoder
Summary
TL;DR
- US Copyright Office rules AI training on copyrighted works is not fair use.
- The agency dismissed industry comparisons between AI and human learning.
- Exceptions for transformative use may be narrowly defined based on purpose.
- Legitimate sourcing of training data is crucial for fair use evaluations.
- Illegally sourced data notably weakens fair use arguments.
Interesting facts
- Generative models are seen as capable of reproducing creative expressions.
- The agency's report indicates a stricter stance on copyright issues in AI.
- Potential exceptions for fair use hinge on specific use cases like research.