security-assessment

This practical guide provides a methodology for assessing the security of AI agent systems, outlining attack classes, controls, and focusing on real-world compromises like data exfiltration or system manipulation.

This paper proposes SEAV, a verification-centric framework for evaluating jailbreak robustness in large language models by assessing response validity and correctness, significantly reducing false-positive rates in safety assessments.

This paper evaluates indirect prompt injection risks in DeepSeek Harness using AI-Infra-Guard for controlled testing, finding notable attack success rates and recommending security controls.