The paper presents AVISE (AI Vulnerability Identification and Security Evaluation), a modular open-source framework designed to identify vulnerabilities and evaluate the security of artificial intelligence systems. With the growing deployment of AI in critical areas, the need for systematic security evaluations is urgent. The authors extend the multi-turn Red Queen attack theory into an Adversarial Language Model (ALM) augmented attack, creating an automated Security Evaluation Test (SET) that includes 25 test cases. This SET demonstrates high accuracy, achieving 92% accuracy, an F1-score of 0.91, and a Matthews correlation coefficient of 0.83, exposing vulnerabilities in nine evaluated language models. AVISE serves as a foundational tool for researchers and industry experts to enhance the rigor and reproducibility of AI security assessments.
Introducing AVISE: A New Framework for Assessing AI System Security
More Articles From This Day
OpenAI Unveils GPT-5.5: The Most Advanced Model for Complex Tasks
OpenAI has announced the release of GPT-5.5, its most advanced language model to date. The new model is designed to be faster and more capable, specifically optimized for complex tasks including coding, research, and data analysis across various tools. This launch reflects OpenAI's commitment to enhancing the functionalities and performance of its AI models.
