Five frontier AI models were forced to reason before making moral choices, and it did not change their final verdicts at all.
Reasoning modes in large language models stabilize ethical outputs without shifting the core binary decision of the machine. Most experts expected that deeper thinking would lead to more nuanced or different moral stances. Instead, the extra processing time merely narrows the gap of disagreement between competing AI models on contentious issues. This suggests that reasoning in AI acts as a consensus-builder toward a universal baseline rather than a tool for moral discovery. For the public, this means AI ethical filters are likely to become more consistent across brands but remain stubbornly fixed in their initial programming.
How Does Thinking Mode Change LLM Moral Judgments? A Controlled Instant-vs-Thinking Comparison Across Five Frontier Models
arXiv · 2605.04488
We evaluate whether enabling provider-exposed reasoning mode changes moral judgments within the same model checkpoint. Across 100 moral-judgment scenarios and five frontier reasoning-trained LLMs (Claude Sonnet 4.6, GPT 5.5, Gemini 3 Flash, DeepSeek V3.1, and Qwen3.5 397B), aggregate binary-verdict agreement remains high and statistically indistinguishable between instant and thinking modes (Krippendorff's alpha = 0.78 vs. 0.79). However, disagreement is concentrated in 21 model-disputed scenari