Outsmarting the Chameleon: Counterfactual Decoupling for Tactical OOD Shifts in Live Streaming Risk Assessment

2026-06-01

AI Summary

This work addresses the challenge of detecting malicious actors in live-streaming scenarios, who evade detection by continuously altering their narrative packaging—termed tactical out-of-distribution (OOD) shifts—while preserving their underlying malicious intent. To tackle this, the paper proposes the LPCD framework, which introduces latent counterfactual disentanglement into live-stream risk control for the first time. LPCD performs causal modeling in latent space to decouple stable malicious intent from variable narrative tactics, anchoring risk predictions via a latent counterfactual consistency constraint. It further incorporates parameter-free inference calibration to mitigate distributional shifts. This approach overcomes the limitations of conventional OOD generalization methods in settings where intent and tactics are tightly coupled, significantly outperforming current state-of-the-art solutions on large-scale industrial datasets and real-world online traffic, thereby enabling robust identification of evolving adversarial risks.

Abstract

Live streaming has emerged as a primary medium for social interaction and digital commerce, yet it is increasingly plagued by sophisticated risks. A fundamental challenge in this domain is \emph{tactical out-of-distribution (OOD) shift}: while malicious actors maintain stable underlying objectives, they continuously redesign narrative packaging to evade detection. Such adversarial shifts expose critical limitations of existing OOD generalization paradigms, whose assumptions are difficult to satisfy in the presence of tightly coupled intent-tactic evolution and ill-defined raw-level counterfactuals. In this paper, we tackle this issue from a \emph{latent causal} perspective and propose \underline{L}atent-\underline{P}redictive \underline{C}ounterfactual \underline{D}ecoupling~(LPCD), a plug-in framework for robust live streaming risk assessment. LPCD enables counterfactual reasoning under adversarial tactical re-packaging by modeling intent and narrative variation at the latent level, and enforces \emph{latent counterfactual consistency} to anchor risk prediction on causally stable malicious intent. At inference time, LPCD applies a lightweight, parameter-free calibration to further mitigate tactic-induced distribution shifts. Extensive experiments on large-scale industrial datasets and online production traffic demonstrate that LPCD consistently outperforms state-of-the-art baselines, validating its effectiveness in moderating evolving adversarial risks in real-world live streaming. The project page is available at https://qiaoyran.github.io/LiveStreamingRiskAssessment/.

Problem

• tactical OOD shift
• live streaming risk assessment
• adversarial evasion
• intent-tactic decoupling
• distribution shift

Innovation

• counterfactual decoupling
• tactical OOD shift
• latent causal modeling
• live streaming risk assessment
• adversarial robustness