REVAL introduces a novel benchmark for evaluating Large Vision-Language Models (LVLMs) across multiple dimensions of reliability and values.

• Assesses security vulnerabilities including adversarial attacks, toxicity, and jailbreak susceptibility
• Evaluates privacy awareness and potential information leakage risks
• Provides a holistic evaluation framework beyond current limited benchmarks
• Offers insights into model performance trade-offs between capabilities and safety

This research is critical for security professionals as it exposes potential weaknesses in vision-language models that could be exploited in real-world applications, enabling more secure AI system development and deployment.

REVAL: A Comprehension Evaluation on Reliability and Values of Large Vision-Language Models