vulnerabilities
Vulnerabilities in AI refer to weaknesses within models or systems that can be exploited, leading to undesirable behavior or outcomes. Identifying and mitigating these vulnerabilities is essential for building robust and trustworthy AI applications.
- A Technical Report on “Erasing the Invisible”: The 2024 NeurIPS Competition on Stress Testing Image Watermarks
- AdvEDM: Fine-grained Adversarial Attack against VLM-based Embodied Agents
- AutoRedTeamer: Autonomous Red Teaming with Lifelong Attack Integration
- BridgePure: Limited Protection Leakage Can Break Black-Box Data Protection
- Enhancing Graph Classification Robustness with Singular Pooling
- LARGO: Latent Adversarial Reflection through Gradient Optimization for Jailbreaking LLMs
- The Curse of Multi-Modalities: Evaluating Hallucinations of Large Multimodal Models across Language, Visual, and Audio
- The Future Unmarked: Watermark Removal in AI-Generated Images via Next-Frame Prediction