Moral Combat: Exploiting Moral Foundations Theory to Jailbreak LLMs

arXiv:2610.19587v1 Announce Type: cross Abstract: Social psychology has long described how people can be persuaded to interpret a harmful act as moral. Recent research suggests that Large Language Models (LLMs) echo aspects of human psychology, from cognitive biases to social conformity. This raises a question with direct safety implications: can LLMs be moralized to bypass safety guardrails? We answer this question with a multi-agent framework grounded in Moral Foundations Theory (MFT). Given a harmful request that is initially refused, our agents diagnose which moral foundation (MF) the prompt is violating, identify a justifiable foundation under which the same request appears virtuous, rewrite the prompt accordingly, and reinforce it with generated precedents of comparable moral reframing. We evaluate our framework on harmful prompts from three well-known jailbreak datasets across five open-source and proprietary models. Our analysis reveals that models are highly susceptible to moralization, exhibiting high single-trial jailbreak rates with success increasing further under repeated attempts. We also find that safety benchmarks are imbalanced in the MFs they violate, concentrating on foundations toward which models already exhibit biases.

LLMs can be manipulated to bypass safety guardrails through moralization, posing a significant risk to safety and security.

This is a concrete red-teaming result showing current LLM safety guardrails are broadly vulnerable to a psychologically grounded manipulation technique, with success rates increasing under repetition. It is also constructive security research that could inform fixes, but as published the immediate implication is a demonstrated, generalizable safety failure.

Researchers demonstrate a novel method to bypass LLM safety guardrails by exploiting moral foundations. This technique successfully moralizes harmful requests, making LLMs interpret them as virtuous, with high and increasing jailbreak rates across various models and datasets. The finding that safety benchmarks are imbalanced in their moral foundations raises further concerns.

The paper demonstrates a method to bypass safety measures in LLMs. This has direct implications for misuse. The success of the method is well-evidenced.

SOURCE     arXiv cs.CY (Computers and Society) (tier 1)
   ↓
DOCUMENT   e8025c6c-2ea7-4e69-9caa-cc0029be7e82
           https://arxiv.org/abs/2610.19587
   ↓
EVIDENCE   2 extracted excerpts
   ↓
MODEL RUN  4 runs, methodology 1.0.1
   ↓
SCORE      -85  (Strongly adverse)
   ↓
CONFIDENCE 85%