Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adversarial Moral Stress Testing of Large Language Models

Saeid Jamshidi, Foutse Khomh, Arghavan Moradi Dakhel, Amin Nikanjam, Mohammad Hamdaqa, Kawser Wazed Nafi

Abstract

Evaluating the ethical robustness of large language models (LLMs) deployed in software systems remains challenging, particularly under sustained adversarial user interaction. Existing safety benchmarks typically rely on single-round evaluations and aggregate metrics, such as toxicity scores and refusal rates, which offer limited visibility into behavioral instability that may arise during realistic multi-turn interactions. As a result, rare but high-impact ethical failures and progressive degradation effects may remain undetected prior to deployment. This paper introduces Adversarial Moral Stress Testing (AMST), a stress-based evaluation framework for assessing ethical robustness under adversarial multi-round interactions. AMST applies structured stress transformations to prompts and evaluates model behavior through distribution-aware robustness metrics that capture variance, tail risk, and temporal behavioral drift across interaction rounds. We evaluate AMST on several state-of-the-art LLMs, including LLaMA-3-8B, GPT-4o, and DeepSeek-v3, using a large set of adversarial scenarios generated under controlled stress conditions. The results demonstrate substantial differences in robustness profiles across models and expose degradation patterns that are not observable under conventional single-round evaluation protocols. In particular, robustness has been shown to depend on distributional stability and tail behavior rather than on average performance alone. Additionally, AMST provides a scalable and model-agnostic stress-testing methodology that enables robustness-aware evaluation and monitoring of LLM-enabled software systems operating in adversarial environments.

Adversarial Moral Stress Testing of Large Language Models

Abstract

Evaluating the ethical robustness of large language models (LLMs) deployed in software systems remains challenging, particularly under sustained adversarial user interaction. Existing safety benchmarks typically rely on single-round evaluations and aggregate metrics, such as toxicity scores and refusal rates, which offer limited visibility into behavioral instability that may arise during realistic multi-turn interactions. As a result, rare but high-impact ethical failures and progressive degradation effects may remain undetected prior to deployment. This paper introduces Adversarial Moral Stress Testing (AMST), a stress-based evaluation framework for assessing ethical robustness under adversarial multi-round interactions. AMST applies structured stress transformations to prompts and evaluates model behavior through distribution-aware robustness metrics that capture variance, tail risk, and temporal behavioral drift across interaction rounds. We evaluate AMST on several state-of-the-art LLMs, including LLaMA-3-8B, GPT-4o, and DeepSeek-v3, using a large set of adversarial scenarios generated under controlled stress conditions. The results demonstrate substantial differences in robustness profiles across models and expose degradation patterns that are not observable under conventional single-round evaluation protocols. In particular, robustness has been shown to depend on distributional stability and tail behavior rather than on average performance alone. Additionally, AMST provides a scalable and model-agnostic stress-testing methodology that enables robustness-aware evaluation and monitoring of LLM-enabled software systems operating in adversarial environments.

Paper Structure

This paper contains 41 sections, 53 equations, 11 figures, 11 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Method: Adversarial Moral Stress Test (AMST)
  4. Problem Definition
  5. Adversarial Stress Transformation
  6. AMST Pipeline
  7. Reasoning Depth Proxy (RDP)
  8. Semantic Ethical Risk (SER)
  9. Refusal Probability (RP)
  10. Moral Deviation Score (MDS)
  11. Robustness Index (RI)
  12. Multi-Round Stress Drift
  13. Threat Model
  14. Theoretical Foundations of Adversarial Moral Stress Testing
  15. Experimental Setup
  16. ...and 26 more sections

Figures (11)

  • Figure 1: Pipeline of the proposed AMST framework.
  • Figure 2: Ethical stability curves illustrating robustness degradation under progressively increasing adversarial stress.
  • Figure 3: Moral drift amplification across sequential adversarial scenarios.
  • Figure 4: Distribution of robustness and moral deviation across different reasoning-depth regimes.
  • Figure 5: Distribution of post-stress robustness conditioned on initial robustness quantiles.
  • ...and 6 more figures