A Comparative Analysis of Ethical and Safety Gaps in LLMs using Relative Danger Coefficient

TL;DR

This work addresses ethical and safety gaps in large language models by introducing the Relative Danger Coefficient ($RDC$), a risk-weighted, penalties-based metric for cross-model comparison. It employs a dual evaluation framework of manual prompts and automated classification to stress-test GPT variants, Gemini series, and DeepSeek-V3(R1) across hazardous information, hate speech, and ethical dilemmas. The RDC aggregates four response categories ($G,U,P,D$) with penalties ($C,S,R,A$) into a 0–100 score, enabling systematic risk profiling. Findings reveal persistent safety vulnerabilities even in reasoning-enabled models, with adversarial framing and complex moral prompts elevating $RDC$ in high-stakes domains, underscoring the need for ongoing human oversight and iterative moderation. The study provides a practical, quantitative basis for benchmarking ethical safety and guiding future improvements in LLM alignment and governance.

Abstract

Artificial Intelligence (AI) and Large Language Models (LLMs) have rapidly evolved in recent years, showcasing remarkable capabilities in natural language understanding and generation. However, these advancements also raise critical ethical questions regarding safety, potential misuse, discrimination and overall societal impact. This article provides a comparative analysis of the ethical performance of various AI models, including the brand new DeepSeek-V3(R1 with reasoning and without), various GPT variants (4o, 3.5 Turbo, 4 Turbo, o1/o3 mini) and Gemini (1.5 flash, 2.0 flash and 2.0 flash exp) and highlights the need for robust human oversight, especially in situations with high stakes. Furthermore, we present a new metric for calculating harm in LLMs called Relative Danger Coefficient (RDC).

Figures (2)

• Figure 1: Model Performance During Manual Testing
• Figure 2: Average Danger Coefficient by Category