AVISE: Framework for Evaluating the Security of AI Systems
Mikko Lempinen, Joni Kemppainen, Niklas Raesalmi
Why It Matters
What makes this one worth your time
As AI systems become integral to critical applications, understanding and mitigating their security vulnerabilities is essential for ensuring their safe deployment.
AVISE offers a new framework for assessing AI security vulnerabilities in language models.
Summary
The paper presents AVISE, a modular open-source framework designed to identify vulnerabilities and evaluate the security of AI systems, demonstrating its effectiveness through an automated Security Evaluation Test that reveals vulnerabilities in language models.
Key contributions
- Introduction of the AVISE framework for AI security evaluation.
- Development of an automated Security Evaluation Test (SET) with 25 test cases.
- Empirical evaluation of nine language models revealing vulnerabilities to the augmented Red Queen attack.
Notable insights
- The extension of the theory-of-mind-based multi-turn Red Queen attack into an Adversarial Language Model augmented attack is a novel approach to AI security evaluation.
- The automated Security Evaluation Test (SET) with high accuracy metrics provides a systematic method for identifying jailbreak vulnerabilities.
Possible limitations
- Not stated in the abstract.
Abstract
arXiv:2604.20833v1 Announce Type: cross Abstract: As artificial intelligence (AI) systems are increasingly deployed across critical domains, their security vulnerabilities pose growing risks of high-profile exploits and consequential system failures. Yet systematic approaches to evaluating AI security remain underdeveloped. In this paper, we introduce AVISE (AI Vulnerability Identification and Security Evaluation), a modular open-source framework for identifying vulnerabilities in and evaluating the security of AI systems and models. As a demonstration of the framework, we extend the theory-of-mind-based multi-turn Red Queen attack into an Adversarial Language Model (ALM) augmented attack and develop an automated Security Evaluation Test (SET) for discovering jailbreak vulnerabilities in language models. The SET comprises 25 test cases and an Evaluation Language Model (ELM) that determines whether each test case was able to jailbreak the target model, achieving 92% accuracy, an F1-score of 0.91, and a Matthews correlation coefficient of 0.83. We evaluate nine recently released language models of diverse sizes with the SET and find that all are vulnerable to the augmented Red Queen attack to varying degrees. AVISE provides researchers and industry practitioners with an extensible foundation for developing and deploying automated SETs, offering a concrete step toward more rigorous and reproducible AI security evaluation.