The US Copyright Office has never granted copyright protection to a work created solely by a machine — not once. That is not a technicality. It is a structural threat to every AI-generated horror series on your platform. Legal scholar Jane Ginsburg at Columbia Law has argued for years that copyright's core requirement is human authorship, and US courts, including the Supreme Court, have held that line consistently. The Ninth Circuit made it explicit in Urantia Foundation v. Maaherra: non-human authored content requires demonstrable human selection to qualify for protection. Last lecture established that the marketing plan is the content plan — horror's native virality does the distribution work. Now the question is: what do you actually own once that content is out there? Copyright law, under the Berne Convention, treats human authors as its foundation. Economic rights — selling, licensing — and moral rights — paternity, integrity — flow from that human origin. AI-generated works disrupt both. The Feist Publications standard requires independent creation plus a minimal degree of creativity; an AI prompt alone may not clear that bar without substantial human creative contribution layered on top. This is where your Creator Agreement becomes load-bearing, Yolanda. Creators on your platform must document their human contribution — the prompt architecture, the editorial choices, the scene sequencing decisions. The UK Copyright, Designs and Patents Act section 9(3) offers a useful model: it deems the person making the necessary arrangements as the author of computer-generated works. That framing protects the human orchestrator. Your agreement should mirror that logic, requiring creators to retain records of their generative process so ownership is defensible if challenged. Training data is the second legal fault line. The New York Times lawsuit against OpenAI put the entire industry on notice: AI trained on copyrighted works without permission raises direct infringement exposure. If creators on your platform use models trained on unlicensed material, that liability can travel upstream to you. Australia's copyright law denies protection outright to non-human created material when human input is low — a signal of where global jurisprudence is trending. The EU AI Act adds a compliance layer: AI providers must maintain documentation, comply with the EU Copyright Directive, and for general-purpose models carrying systemic risk, conduct red-teaming and cybersecurity evaluations. Ethical risks compound the legal ones — systemic biases in race, gender, and sexuality embedded in training data can surface in your content, creating both reputational and regulatory exposure. Content moderation is where legal compliance meets platform survival, Yolanda. App stores — Apple and Google — enforce content guidelines that can delist your app without warning. AI assists moderation by flagging prohibited content at scale, but the EU AI Act's prohibition on certain biometric identification and social scoring practices means your moderation pipeline itself must be audited for compliance. The so-called black box problem in deep learning directly conflicts with the Right to Explanation embedded in EU regulation — if your AI moderation system cannot explain a removal decision, you face both legal and creator-relations risk. Trustworthy AI, per the EU framework, requires seven properties: human agency, oversight, technical robustness, privacy, transparency, fairness, and accountability. Build those into your moderation architecture from day one, not as a retrofit.