Multimodal UGC Moderation and Compliance Pipeline

You are the trust-and-safety architect for [INSERT GAME TITLE], a global UGC platform where players upload levels, assets, text, audio, and live behavior. Design a multimodal AI moderation pipeline enforcing safety, ratings, privacy, IP, and regional law without blocking legitimate creativity. Module 1 - Gameplay design: map community guidelines to concrete content actions, define age-rating bands, appeal expectations, and the creator experience during review. Module 2 - Generation pipeline: specify classifiers for text, images, 3D meshes, audio, and code; PII detection; IP-similarity matching; risk scoring; human routing; and real-time chat/voice filters. Include pre-publish scoring, post-publish re-scan as models update, and hash sharing with industry safety consortia. Define blocking, asynchronous, and near-real-time paths with distinct thresholds and SLAs. Address generative risks: prompt injection inside level data, steganographic imagery, and code-as-content exploits. Define actionable creator feedback for self-correction before resubmission. Specify retention and deletion rules for scanned content plus model-training opt-outs. Module 3 - Data structures: provide schemas for review items, verdicts, policy versions, appeals, reporter signals, and explainability records. Module 4 - Testing: define cross-language red-team suites, false-positive creator studies, adversarial evasion tests, and drift monitoring. Module 5 - Operations: cover model update governance, policy localization for [INSERT REGIONS], auditor access, transparency reports, appeal SLAs, and incident runbooks. Provide a text architecture diagram, decision-flow diagram, risk register, and KPIs for precision, recall, appeal overturn rate, and time-to-action. List assumptions about scale, hosting jurisdictions, and rating frameworks. Begin with a one-page safety model summary, then deliver modules one at a time for approval. ================