AI copyright defense may fail under fruit of poisonous tree doctrine

A legal theory gaining traction argues that AI-generated works cannot receive copyright protection under the fruit of the poisonous tree doctrine applied to intellectual property. Legal commentator Trev Baylis asserts that even edited AI outputs or those with thin copyright for selection and arrangement may fail under Section 103(a) and the Anderson v. Stallone standard if training data is deemed infringing. This interpretation suggests individual AI users would face the burden of proving an offensive fair use defense rather than relying on derivative work protections. Courts have not yet settled whether this criminal law concept applies to copyright eligibility for generative AI. The argument challenges current industry assumptions that significant human modification creates new protectable expression. If validated judicially, this framework could invalidate copyright claims across the generative AI sector by linking output eligibility directly to upstream data licensing compliance.