Policy-Grounded Safety Evaluation of 20 Large Language Models

TL;DR

This paper presents Aymara AI, a platform for scalable, policy-grounded safety evaluation of 20 large language models across diverse real-world domains, revealing significant performance disparities and emphasizing safety assessment challenges.

Contribution

Introduction of Aymara AI, a novel platform that transforms safety policies into adversarial prompts and scores models, enabling comprehensive safety evaluation across multiple LLMs.

Findings

Models scored highest in misinformation (mean 95.7%)

Models performed poorly in privacy and impersonation (mean 24.3%)

Safety scores varied significantly across models and domains (p < .05)

Abstract

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.