Think Before You Lie: How Reasoning Improves Honesty

TL;DR

The effect of reasoning is interpreted: generating deliberative tokens as part of moral reasoning entails the traversal of a biased representational space, ultimately nudging the model toward its more stable, honest defaults.

Abstract

While existing evaluations of large language models (LLMs) measure deception rates, the underlying conditions that give rise to deceptive behavior are poorly understood. We investigate this question using a novel dataset of realistic moral trade-offs where honesty incurs variable costs. Contrary to humans, who tend to become less honest given time to deliberate (Capraro, 2017; Capraro et al., 2019), we find that reasoning consistently increases honesty across scales and for several LLM families. This effect is not only a function of the reasoning content, as reasoning traces are often poor predictors of final behaviors. Rather, we show that the underlying geometry of the representational space itself contributes to the effect. Namely, we observe that deceptive regions within this space are metastable: deceptive answers are more easily destabilized by input paraphrasing, output resampling, and activation noise than honest ones. We interpret the effect of reasoning in this vein: generating deliberative tokens as part of moral reasoning entails the traversal of a biased representational space, ultimately nudging the model toward its more stable, honest defaults.

Figures (23)

• Figure 1: Experimental setup.
• Figure 2: Schematic illustration of the answer space: honesty occupies a larger region than deception, and fewer paths lead to deception.
• Figure 3: Percentage of the time reasoning improved the probability of honesty (blue bar), % of the time reasoning decreased the probability of honesty (red bar), % of cases where token-forcing leads to deception but reasoning leads to honesty (tick mark inside blue bar), % of cases where token-forcing leads to honesty but reasoning leads to deception (tick mark inside red bar). Effects are shown for different sentence lengths constraints. "X" means no constraint was provided. See App. Figure \ref{['fig:reasoning_effect_probabilities']} for how effect sizes translate to differences in the probability of honesty, and see App. Figure \ref{['fig:reasoning_effect_probabilities_thinking']} for results on thinking models.
• Figure 4: Left: Flip rates after either resampling or paraphrasing grouped by the polarity (whether deceptive or honest) of the baseline. Flip rates are much higher for the deceptive group, which means that changing seeds and paraphrases is much likelier to flip deceptive answers than honest ones. Right: Flip rates given activation noise in reasoning mode. Experiments are run on $3$ seeds with $m_{fraction}=0.02$. Noise is applied over all decoding steps. For both DoubleBind and DailyDilemmas, adding noise is significantly more likely to flip deceptive answers than honest ones.
• Figure 5: Left: Average segment length over the course of reasoning for gemma-3-4b-it on DailyDilemmas when token-forced and post-reasoning responses diverge, and average flip rates for the same set. Deceptive segments are shorter and more frequently flip to honesty over the course of reasoning, indicating less stability. Right: Pairwise similarity between honest and dishonest pairs for gemma-3-12b-it. Dishonest pairs are less similar (0.88) than honest pairs (0.94). See more figures in App. \ref{['app:hh-dd:cos']}