Automated Safety Evaluations Across 20 Large Language Models: The Aymara LLM Risk and Responsibility Matrix
This addresses the need for rigorous safety assessments in real-world AI applications, though it is incremental as it builds on existing evaluation methods with a new tool.
The paper tackled the problem of scalable safety evaluation for large language models (LLMs) by introducing Aymara AI, a platform that transforms safety policies into adversarial prompts and scores model responses, resulting in evaluations of 20 LLMs across 10 domains with mean safety scores ranging from 86.2% to 52.4%.
As large language models (LLMs) become increasingly integrated into real-world applications, scalable and rigorous safety evaluation is essential. This paper introduces Aymara AI, a programmatic platform for generating and administering customized, policy-grounded safety evaluations. Aymara AI transforms natural-language safety policies into adversarial prompts and scores model responses using an AI-based rater validated against human judgments. We demonstrate its capabilities through the Aymara LLM Risk and Responsibility Matrix, which evaluates 20 commercially available LLMs across 10 real-world safety domains. Results reveal wide performance disparities, with mean safety scores ranging from 86.2% to 52.4%. While models performed well in well-established safety domains such as Misinformation (mean = 95.7%), they consistently failed in more complex or underspecified domains, notably Privacy & Impersonation (mean = 24.3%). Analyses of Variance confirmed that safety scores differed significantly across both models and domains (p < .05). These findings underscore the inconsistent and context-dependent nature of LLM safety and highlight the need for scalable, customizable tools like Aymara AI to support responsible AI development and oversight.