Think Before You Lie: How Reasoning Leads to Honesty

Ann Yuan; Asma Ghandeharioun; Carter Blum; Alicia Machado; Jessica Hoffmann; Daphne Ippolito; Martin Wattenberg; Lucas Dixon

cs.AIMar 10, 2026

Think Before You Lie: How Reasoning Leads to Honesty

Authors: Ann Yuan, Asma Ghandeharioun, Carter Blum, Alicia Machado, Jessica Hoffmann, Daphne Ippolito, Martin Wattenberg, Lucas Dixon,

Organizations: Google DeepMind · Carnegie Mellon University · Harvard University

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 & tables

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. 2 2 2 Schematic built using https://apps.amandaghassaei.com/gpu-io/examples/fluid/

Panel of Figure 3 — 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 7 for how effect sizes translate to differences in the probability of honesty, and see App. Figure 8 for results on thinking models.

Panel of Figure 4 — 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.

Panel of Figure 5 — 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. A.10

Noise	Type	Mean Prob	Survival ( $>0.5$ )
Qwen 3 4B Instruct
0.0	H-H	1.00 ±0.00	100.0% (192/192)
0.0	D-D	1.00 ±0.01	100.0% (192/192)
0.8	H-H	0.96 ±0.15	93.2% (179/192)
0.8	D-D	0.94 ±0.19	90.1% (173/192)
1.0	H-H	0.92 ±0.21	89.6% (172/192)

Table 1: Last layer hidden representation interpolation across honest (deceptive) trajectories. Noise coefficient is further multiplied with the mean norm of the start and end points

Figure 6: Reasoning has an asymmetric effect on the probability of honest outcomes. If the deceptive option is listed last, reasoning more significantly improves the probability of honesty.

Appendix figures & tables21 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 7 — Figure 7: The difference in the probability of the token representing honesty between token forcing and reasoning for various sentence counts. Blue bars means reasoning yielded a higher probability of honesty.

Panel of Figure 8 — Figure 8: Results for thinking models. The two plots on the left show difference in probability of the token representing honesty between token forcing and reasoning for various token counts. The two plots on the right show the percentage of cases where reasoning increased honesty (blue bars) versus decreased honesty.

Panel of Figure 9 — Figure 9: The probability of the token representing honesty.

Panel of Figure 10 — Figure 10: The probability of the token representing honesty for thinking models.

Panel of Figure 11 — Figure 11: Effects of reasoning on honesty for Gemini 3 Flash. Reasoning does improve honesty, though the effect of the reasoning budget is negligible.

Figure 12: Overlap between models in terms of which scenarios receive an honesty boost after reasoning, as measured by the intersection over union (Jaccard index).

Figure 13: DoubleBind dataset - Deception stability under input paraphrasing and resampling

Figure 14: DailyDilemmas dataset - Deception stability under input paraphrasing and resampling

Figure 15: Average segment length over the course of reasoning for gemma-3-4b-it on DailyDilemmas dataset when token forcing and reasoning responses diverge.

Figure 16: Given scenarios where reasoning changed the answer relative to token-forcing, we token-force an answer at each sentence boundary within the reasoning chain to understand when the model converges on a final decision (“Answer Stabilization”), when the model first ventures what will become its final decision (“Answer Discovery”), how long the overall reasoning traces tend to be (“Total Length”), and how often merely being instructed to reason is enough to flip the model’s answer (“First Index Equals 0”).

Figure 17: More reasoning generally reduces recency bias.

Model	Budget	CoT Order	Win rate (%)
gemma-12b-it	2,048 tokens	H — D	78
		D — H	72
		both	75
qwen-4b-instruct	2,048 tokens	H — D	73.5
		D — H	72.9
		both	73.2

Appendix

Table 2: Linearity scores across models and reasoning budgets. The numbers show the win rate of the honest versus deceptive Chain-of-Thought (CoT). H-D refers to the examples where the honest CoT was listed first, and D-H refers to the ones where the deceptive CoT was listed first in the prompt given to the autorater.

Noise	Type	Mean Prob	Survival Rate ( $>0.5$ )
0.0	H-H	0.98 ±0.01	100.0% (294/294)
0.0	D-D	0.66 ±0.33	49.7% (146/294)
0.8	H-H	0.92 ±0.11	95.9% (282/294)
0.8	D-D	0.59 ±0.34	43.9% (129/294)
1.0	H-H	0.88 ±0.19	87.4% (257/294)
1.0	D-D	0.54 ±0.35	38.4% (113/294)

Appendix

Table 3: Last layer hidden representation interpolation across honest (deceptive) trajectories (Olmo 3 7B). Noise coefficient is further multiplied with the mean norm of the start and end points

Noise	Type	Mean Prob	Survival Rate ( $>0.5$ )
0.0	H-H	1.00 ±0.00	100.0% (315/315)
0.0	D-D	1.00 ±0.00	100.0% (315/315)
0.8	H-H	1.00 ±0.00	100.0% (315/315)
0.8	D-D	1.00 ±0.02	100.0% (315/315)
1.0	H-H	1.00 ±0.01	99.7% (314/315)
1.0	D-D	0.99 ±0.04	98.7% (311/315)

Appendix

Table 4: Last layer hidden representation interpolation across honest (deceptive) trajectories (Gemma 3 4B). Noise coefficient is further multiplied with the mean norm of the start and end points

Figure 18: Olmo 3 7b. Vector Space Compactness: How similar are the CoTs? Pairwise similarity between dishonest pairs is lower than honest pairs: Honest-Honest: 0.96, Deceptive-Deceptive: 0.87.

Figure 19: Gemma 3 4b. Vector Space Compactness: How similar are the CoTs? Pairwise similarity between dishonest pairs is comparable with that of honest pairs: Honest-Honest: 0.95, Deceptive-Deceptive: 0.94.

Figure 20: Qwen 3 4b. Vector Space Compactness: How similar are the CoTs? Pairwise similarity between dishonest and honest pairs is very high. Honest-Honest: 0.99, Deceptive-Deceptive: 0.99.

Figure 21: Gemma family of models demonstrate elasticity with respect to costs: as costs increase, deception propensity also increases. Rising cost indices are plotted along the x-axis of each chart. Charts are clustered by scenario, with larger models on the right. The probability of the honest token (when token forcing) is represented by the blue line, and the probability of the deceptive token is represented by the red line. 95% confidence intervals are displayed.

Figure 22: Flip rates for token forcing setup. Noise level is at 2% of the norm of the last hidden state in the transformer, and this process is applied once before the token prediction. The results are from running with $5$ different seeds. We present results in both Double Bind and Daily Dilemmas datasets and we observe that adding noise has no significant effect for the token forcing setup. We only see significant results for Gemma 3 12B IT on DailyDilemmas .

Model	Layer
gemma-4b-it	$20$
gemma-12b-it	$25$
gemma-27b-it	$35$
qwen-4b-instruct	$20$
qwen-30b-instruct	$25$
olmo-7b-instruct	$20$

Appendix

Table 5: Layers used for additive noise experiments.

Panel of Figure 23 — Figure 23: Various angles of a 3-D PCA of embeddings along reasoning trajectories. The honest embeddings are in blue, the deceptive ones in orange. Blue (honest) embeddings seem to be more widespread and occupy a larger region of space, while orange (deceptive) embeddings are more localized.

Deception in Language Models

Deception in language models studies when systems mislead through lies, omissions, or selective truths, and how to detect and prevent it.

See all 97 papers

When Do Large Language Models Exhibit Unsolicited Deception?
Mar 31, 2025Samuel M. Taylor, Benjamin K. BergenLanguage Model Safety Evaluation Deception in Language Models
Department of Cognitive Science University of California, San Diego
Confidently Deceptive: On the Relationship Between Confidence and Deception in LLMs
May 12, 2026Ali Asad, Stephen Obadinma, Anshul Pattoo +2Language Model Safety Evaluation Confidence Estimation in Language Models
Department of Electrical and Computer Engineering & Ingenuity Labs Research Institute Queen’s University, Kingston, Canada
Reasoning-Token Spikes Under Prompted Untruthful Responses in Large Language Models
Oct 7, 2026Maverick Morales, Tomáš Dominik, Vermut Gao +4Deception in Language Models LLM Auditing
Chapman University, CA, USA · INSERM U992, Cognitive Neuroimaging Unit, NeuroSpin Center, Gif sur Yvette 91191, France · University of California, Los Angeles, CA, USA +1

Mar 31, 2025J/K move · Enter open · S save

When Do Large Language Models Exhibit Unsolicited Deception?

Samuel M. Taylor, Benjamin K. Bergen

Department of Cognitive Science University of California, San Diego

Language Model Safety Evaluation Deception in Language Models

Open paper AskPDF arXiv

Large Language Models (LLMs) are effective at deceiving when prompted to do so. Models that demonstrate better performance on reasoning tasks are also better at prompted deception. But under what conditions do they deceive without instruction to do so? This study evaluates unsolicited deception produced by LLMs in a preregistered experimental protocol using tools from signaling theory. We evaluated a range of 18 proprietary closed-source and open-source LLMs using modified 2x2 games (in the style of the Prisoner's Dilemma) augmented with a phase in which they can freely communicate to the other agent using unconstrained language. This setup creates an opportunity to misrepresent its actions in conditions that vary in how useful doing so might be towards goal satisfaction. The results indicate that 1) all tested LLMs misrepresent their actions in at least some conditions, 2) they are generally more likely to do so in situations in which deception is beneficial, and 3) models exhibiting better reasoning capacity overall tend to misrepresent at higher rates. Taken together, these results suggest a correlational relationship between model reasoning performance and situational deception, and reveal certain contextual factors that affect whether LLMs will misrepresent actions or not in a novel experimental configuration.

Abstract

Deception in Language Models

Explore similar work

When Do Large Language Models Exhibit Unsolicited Deception?

Confidently Deceptive: On the Relationship Between Confidence and Deception in LLMs

Reasoning-Token Spikes Under Prompted Untruthful Responses in Large Language Models