Fengbin Zhu

This paper introduces the concept of 'Sleeper Attack,' demonstrating that adversarial content can persist across multiple interactions with an LLM agent, posing a more subtle and difficult-to-detect safety threat than single-interaction attacks.

The paper proposes the Shortcut Subspace Suppression (S^3) framework to improve deepfake detection generalization by explicitly identifying and suppressing method-specific shortcuts in learned feature representations.