We posted an article recently covering the Copyright Office’s report on AI training and fair use.

Just to review quickly, Fair Use is the provision of copyright law that allows for limited usage of copyrighted materials in certain conditions without needing to obtain prior permission from the copyright holder.

AI “training” refers to the specific data, including copyrighted works, from which the AI learns.

A recent lawsuit, filed by thirteen authors, accused Meta of copyright infringement. The judge in the case ruled that the plaintiffs had failed to show that Meta’s models produced outputs substantially similar to their copyrighted works. He noted that it “is generally illegal to copy protected works without permission” but found that they had not demonstrated a compelling argument that Meta’s use of the material caused market harm.

It should be noted that the court ruled in this instance that the usage of copyrighted material counted as fair use, but only because there was insufficient evidence of market harm. This was not a ruling on the broader issue of AI training. There is a separate claim that Meta may have unlawfully distributed the works, which was not dismissed and will proceed.

The judge rejected Meta’s argument that public interest justified unauthorized use of copyrighted material. Meta’s claim that barring usage of copyrighted material would halt AI development was called “nonsense”, and the judge added that if the books were as valuable to AI as Meta claims, they should find a way to license them.

While Meta won this particular case, the judge noted that stronger claims and better evidence could render a different outcome.