Anthropic CEO Dario Amodei strongly defends his company's use of copyrighted materials in AI training, asserting it qualifies as fair use under current law.

Speaking to Ezra Klein of The New York Times, Amodei explained that their AI models don't simply reproduce content but transform it in a way similar to human learning. "I think everyone agrees the models shouldn't be verbatim outputting copyrighted content," he stated, emphasizing that the training process is "sufficiently transformative" to qualify as fair use.

This stance comes amid ongoing legal challenges. In 2023, major music publishers Concord, Universal, and ABKCO filed a lawsuit against Anthropic in Tennessee, alleging copyright infringement related to song lyrics used in training their AI chatbot, Claude.

Anthropic's legal defense rests on several key arguments:

• The use of lyrics is transformative and adds new purpose to the original works
• Song lyrics comprise a "miniscule fraction" of their training data
• The scale of licensing required would be practically impossible
• The publishers themselves prompted Claude to produce the infringing content

The company maintains that their training methods are fundamentally different from simple reproduction, arguing that the transformation of content through AI training constitutes fair use under current law. This legal battle represents a crucial test case for AI companies' use of copyrighted materials in training large language models.

Related Articles

Previous Articles