SATBA: An Invisible Backdoor Attack Based on Spatial Attention
SATBA proposes an imperceptible backdoor attack using spatial attention, overcoming limitations of existing methods and ensuring high attack success rate.
ImgTrojan: Jailbreaking Vision-Language Models with ONE Image
VLMs are vulnerable to ImgTrojan attacks, compromising safety barriers with poisoned images.
Gradient Cuff: Detecting Jailbreak Attacks on Large Language Models
Gradient Cuff proposes a two-step method to detect jailbreak attempts on Large Language Models by exploring refusal loss landscapes.
DrAttack: Prompt Decomposition and Reconstruction for Effective LLM Jailbreaking
Decomposing and reconstructing prompts can effectively jailbreak LLMs, concealing malicious intent and increasing success rates.
Auditable Homomorphic-based Decentralized Collaborative AI with Attribute-based Differential Privacy
Proposing AerisAI for secure decentralized AI collaboration with differential privacy and homomorphic encryption.
LLM Jailbreaking Defense via Backtranslation
Proposing a defense method using backtranslation to protect LLMs from jailbreaking attacks.
Practical Threat Models in AI Security: Bridging Research and Practice
Bridging the gap between academic threat models and practical AI security by studying real-world threat models.
Analysis of Transfer Attack on Image Watermarks
Watermark-based AI-generated image detectors are vulnerable to transfer evasion attacks, even in the absence of access to detection APIs.
Bergeron: A Framework for AI Model Protection Against Adversarial Attacks
Bergeron introduces a framework to enhance the robustness of AI models against adversarial attacks without additional training.
Defending Object Detection AI Models Against Adversarial Patch Attacks
Robust defense strategies are crucial in countering adversarial patch attacks on object detection AI models.
© 2024 by Linnk AI