Netflix issues GenAI rules:no generative AI in final deliverables (simonwillison.net)

Netflix has published a set of GenAI guidelines for production partners that effectively ban generative outputs from being included in final deliverables unless specific safeguards are met. Key rules require that outputs not replicate or substantially recreate identifiable characteristics of unowned or copyrighted material, that generative tools must not store, reuse, or train on production inputs/outputs, and that AI use happen in enterprise‑secured environments. Generated material must remain temporary (not part of final assets) and GenAI must not replace or create performances or other union‑covered work without consent. Partners who can’t answer “yes” to these principles must escalate to their Netflix contact — written approval may be required. For the AI/ML community this crystallizes production‑grade requirements around IP, data governance, and labor rules. Technically, it favors ephemeral inference-only workflows, on‑prem or VPC‑isolated models, strict logging/retention controls, and bans fine‑tuning on proprietary production data unless explicitly allowed. Toolmakers and vendors will need to demonstrate non‑training guarantees, secure enterprise deployments, and provenance controls to integrate into studio pipelines. The guidance signals stronger enterprise demand for models and platforms that provide auditable data handling, configurable retention policies, and explicit consent controls for talent and union considerations.