The Indirect Disclosure Effect: How Disclosing Generative AI Use Impacts Human Creative Collaboration with AI

Generative AI disclosure rules aim to protect audiences from deception, preserving human creativity and self-expression. Yet disclosure may change not only how audiences evaluate creative work but also how creators produce it. In two experiments involving collaboration with a text-to-image generative AI tool, we identify an indirect disclosure effect. When creators anticipated that their AI use would be disclosed to a lay audience, most withdrew from the creative process and gave the AI greater control. They did so because they feared that audiences would discount their human creative agency and fail to recognize the work as an expression of their creative selves. The resulting images reflected more computational than human creativity: audiences viewed them as more novel but less appealing and lower in quality. This withdrawal did not occur when creators expected evaluation by experts who could better recognize human creative agency in AI-assisted work. Our findings reveal a policy tension wherein simple AI-use labels can inadvertently weaken the human agency that transparency rules seek to protect. Regulators, platforms, and organizations should, therefore, complement disclosure requirements with audience AI literacy and more informative, process-based disclosures that show how people contributed through prompting, selection, revision, and editing.