An Adaptive, Generative Framework for Continuous Multimodal Avatar Authentication in the Metaverse

Source abstract

Metaverse technologies let users interact inside virtual environments, where each user is represented by an avatar that carries their digital identity. These environments now attract many application domains, and these domains do not all need the same level of security. Conventional authentication methods do not fit them well. They ignore the continuity of user activity during a session and do not exploit the metaverse's rich set of interaction modalities, far broader than in 2D systems, through which an avatar continuously produces voice, head movement, gaze, and others. Moreover, they cannot adjust to different security levels within the same environment. Therefore, continuous, non-intrusive and adaptive authentication is needed. We first review existing avatar authentication approaches, organize them into three categories, and draw out the limitations a solution must address. Building on this, we propose an AI-driven continuous multimodal authentication framework for the metaverse. The framework reads behavioral biometrics from several interaction modalities and authenticates the user through their avatar without breaking immersion. The security level is continuously adjusted to the context of the session, using an adaptive control mechanism that weighs scene sensitivity, risk, and trust. A generative AI component builds context-aware authentication scenarios that adjust to the required security level and to the sensitivity of the scene. We evaluate the framework in two ways: a comparison against existing work, and a scenario-based analysis. The comparison shows that our framework integrates capabilities that previous approaches generally address separately, while the scenario-based analysis illustrates its intended operation and its internal consistency with respect to adaptivity and context-awareness.