Study Finds LLM Moral Choices Lack Stable Value Rankings

1 reports archived1 independent sourcesupdated 11/11/2026, 12:08:45

Synthesis & Latest Updates

1 Sources Cross-Validated

A study of five LLMs including GPT-5.2, Gemini and DeepSeek found that models lack a consistent value ranking when comparing pairs of moral violations: reversing option order changed the ranking in 26 of 30 test blocks. Critically, averaging responses produces a clean, human-like moral hierarchy that matches no individual test set, so moral alignment scores should only be trusted if rankings hold on individual prompts and both answer orders.

LATEST/A study of five LLMs including GPT-5.2, Gemini and DeepSeek found that models lack a consistent value ranking when comparing pairs of moral violations: reversing option order changed the ranking in 26 of 30 test blocks. Critically, averaging responses produces a clean, human-like moral hierarchy that matches no individual test set, so moral alignment scores should only be trusted if rankings hold on individual prompts and both answer orders.

Coverage timeline

Total 1 reports · Latest first

Rohan Paul (@rohanpaul_ai)T2Nov 11, 12:08·71 pts
Study Finds LLM Moral Choices Lack Stable Value Rankings
- Reversing the order of two moral-violation options changed the value ranking in 26 of 30 test blocks, indicating unstable model choices.
- Averaging responses yields a clean, human-like moral hierarchy that matches no individual test set.
- Before trusting moral alignment scores, verify consistent rankings on individual prompts and both answer orders rather than relying on averages.
View Related Reports & Analysis Original Source