The Fragility Of Moral Judgment In Large Language Models

Researchers tested the stability of moral judgments in large language models using nearly 3,000 ethical dilemmas, finding that narrative framing and evaluation methods significantly influence AI decisions. The study reveals that LLM moral reasoning is highly dependent on how questions are presented rather than underlying moral substance, with only 35.7% consistency across different evaluation protocols.

• →Surface text changes caused minimal judgment shifts (7.5%), but perspective changes led to 24.3% judgment reversals in AI models.
• →Only 67.6% agreement exists between different evaluation protocols for the same moral dilemmas across four major LLMs.
• →Morally ambiguous scenarios where no clear blame exists are most susceptible to judgment manipulation.
• →Persuasion techniques can systematically bias AI moral decisions in predictable directions.
• →The findings raise concerns about AI reliability for moral guidance as outcomes depend more on presentation skills than ethical substance.