Consent-Aware Multimodal Moderation for Social VR

Social VR harassment is emergent and contextual (Schulenberg et al., 2023), and ethics work flags the need for hybrid human–AI approaches with strong privacy and governance (Zhuk, 2024). At the same time, multimodal LLMs show promise in complex video/text tasks but have notable failure cases (Levi et al., 2025). We propose a consent-aware architecture where on-device multimodal models analyze proximity, voice tone, gaze/gesture vectors, and repetition patterns to detect harassment trajectories—not just single events. Users have real-time consent controls (personal safety radius, “do-not-engage” states) that AI enforces locally. Escalation to human moderators occurs when patterns persist or cross severity thresholds, with privacy-preserving evidence summaries. Conditional delegation (Lai et al., 2022) governs which signals the AI can act on autonomously vs. which require human review. Novelty lies in combining trajectory-aware detection, user-controlled consent signals, and privacy-by-design evidence generation, addressing both safety and autonomy in a modality-rich environment. The result is a more nuanced, user-centered moderation model for immersive spaces.

References:

1. Human-AI Collaboration via Conditional Delegation: A Case Study of Content Moderation. Vivian Lai, Samuel Carton, Rajat Bhatnagar, Vera Liao, Yunfeng Zhang, Chenhao Tan, Q. Liao (2022). International Conference on Human Factors in Computing Systems.
2. Towards Leveraging AI-based Moderation to Address Emergent Harassment in Social Virtual Reality. Kelsea Schulenberg, Lingyuan Li, Guo Freeman, Samaneh Zamanifard, Nathan J. Mcneese (2023). International Conference on Human Factors in Computing Systems.
3. Ethical implications of AI in the Metaverse. A. Zhuk (2024). AI and Ethics.
4. AI vs. Human Moderators: A Comparative Evaluation of Multimodal LLMs in Content Moderation for Brand Safety. Adi Levi, Or Levi, Sardhendu Mishra, Jonathan Morra (2025). arXiv.org.