Evaluating & Reducing Deceptive Dialogue From Language Models with Multi-turn RL

• URL: http://arxiv.org/abs/2510.14318v1
• Date: Thu, 16 Oct 2025 05:29:36 GMT
• Title: Evaluating & Reducing Deceptive Dialogue From Language Models with Multi-turn RL
• Authors: Marwa Abdulhai, Ryan Cheng, Aryansh Shrivastava, Natasha Jaques, Yarin Gal, Sergey Levine,
• Abstract summary: Large Language Models (LLMs) interact with millions of people worldwide in applications such as customer support, education and healthcare.<n>Their ability to produce deceptive outputs, whether intentionally or inadvertently, poses significant safety concerns.<n>We investigate the extent to which LLMs engage in deception within dialogue, and propose the belief misalignment metric to quantify deception.
• Score: 64.3268313484078
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models (LLMs) interact with millions of people worldwide in applications such as customer support, education and healthcare. However, their ability to produce deceptive outputs, whether intentionally or inadvertently, poses significant safety concerns. The unpredictable nature of LLM behavior, combined with insufficient safeguards against hallucination, misinformation, and user manipulation, makes their misuse a serious, real-world risk. In this paper, we investigate the extent to which LLMs engage in deception within dialogue, and propose the belief misalignment metric to quantify deception. We evaluate deception across four distinct dialogue scenarios, using five established deception detection metrics and our proposed metric. Our findings reveal this novel deception measure correlates more closely with human judgments than any existing metrics we test. Additionally, our benchmarking of eight state-of-the-art models indicates that LLMs naturally exhibit deceptive behavior in approximately 26% of dialogue turns, even when prompted with seemingly benign objectives. When prompted to deceive, LLMs are capable of increasing deceptiveness by as much as 31% relative to baselines. Unexpectedly, models trained with RLHF, the predominant approach for ensuring the safety of widely-deployed LLMs, still exhibit deception at a rate of 43% on average. Given that deception in dialogue is a behavior that develops over an interaction history, its effective evaluation and mitigation necessitates moving beyond single-utterance analyses. We introduce a multi-turn reinforcement learning methodology to fine-tune LLMs to reduce deceptive behaviors, leading to a 77.6% reduction compared to other instruction-tuned models.

Related papers

• Vulnerability of LLMs' Belief Systems? LLMs Belief Resistance Check Through Strategic Persuasive Conversation Interventions [8.026492468995187]
  Small models exhibit extreme compliance, with over 80% of belief changes occurring at the first persuasive turn.<n> meta-cognition prompting increases vulnerability by accelerating belief erosion rather than enhancing robustness.<n>These findings highlight substantial model-dependent limits of current robustness interventions.
  arXiv  Detail & Related papers  (2026-01-20T04:43:55Z)
• PerProb: Indirectly Evaluating Memorization in Large Language Models [13.905375956316632]
  We propose PerProb, a label-free framework for indirectly assessing LLM vulnerabilities.<n>PerProb evaluates changes in perplexity and average log probability between data generated by victim and adversary models.<n>We evaluate PerProb's effectiveness across five datasets, revealing varying memory behaviors and privacy risks.
  arXiv  Detail & Related papers  (2025-12-16T17:10:01Z)
• Beyond Prompt-Induced Lies: Investigating LLM Deception on Benign Prompts [79.1081247754018]
  Large Language Models (LLMs) are widely deployed in reasoning, planning, and decision-making tasks.<n>We propose a framework based on Contact Searching Questions(CSQ) to quantify the likelihood of deception.
  arXiv  Detail & Related papers  (2025-08-08T14:46:35Z)
• Teaching Language Models To Gather Information Proactively [53.85419549904644]
  Large language models (LLMs) are increasingly expected to function as collaborative partners.<n>In this work, we introduce a new task paradigm: proactive information gathering.<n>We design a scalable framework that generates partially specified, real-world tasks, masking key information.<n>Within this setup, our core innovation is a reinforcement finetuning strategy that rewards questions that elicit genuinely new, implicit user information.
  arXiv  Detail & Related papers  (2025-07-28T23:50:09Z)
• How Much Content Do LLMs Generate That Induces Cognitive Bias in Users? [13.872175096831343]
  Large language models (LLMs) are increasingly integrated into applications ranging from review summarization to medical diagnosis support.<n>We investigate when and how LLMs expose users to biased content and quantify its severity.<n>Our findings show that LLMs expose users to content that changes the sentiment of the context in 21.86% of the cases, hallucinates on post-knowledge-cutoff data questions in 57.33% of the cases, and primacy bias in 5.94% of the cases.
  arXiv  Detail & Related papers  (2025-07-03T21:56:44Z)
• MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs [66.14178164421794]
  We introduce MetaFaith, a novel prompt-based calibration approach inspired by human metacognition.<n>We show that MetaFaith robustly improves faithful calibration across diverse models and task domains, enabling up to 61% improvement in faithfulness.
  arXiv  Detail & Related papers  (2025-05-30T17:54:08Z)
• Robots in the Middle: Evaluating LLMs in Dispute Resolution [0.0]
  We investigate whether large language models (LLMs) are able to analyze dispute conversations, select suitable intervention types, and generate appropriate intervention messages. Our results demonstrate the potential of integrating AI in online dispute resolution (ODR) platforms.
  arXiv  Detail & Related papers  (2024-10-09T16:51:10Z)
• Rel-A.I.: An Interaction-Centered Approach To Measuring Human-LM Reliance [73.19687314438133]
  We study how reliance is affected by contextual features of an interaction. We find that contextual characteristics significantly affect human reliance behavior. Our results show that calibration and language quality alone are insufficient in evaluating the risks of human-LM interactions.
  arXiv  Detail & Related papers  (2024-07-10T18:00:05Z)
• SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal [64.9938658716425]
  SORRY-Bench is a proposed benchmark for evaluating large language models' (LLMs) ability to recognize and reject unsafe user requests.<n>First, existing methods often use coarse-grained taxonomy of unsafe topics, and are over-representing some fine-grained topics.<n>Second, linguistic characteristics and formatting of prompts are often overlooked, like different languages, dialects, and more -- which are only implicitly considered in many evaluations.
  arXiv  Detail & Related papers  (2024-06-20T17:56:07Z)
• Prompt Leakage effect and defense strategies for multi-turn LLM interactions [95.33778028192593]
  Leakage of system prompts may compromise intellectual property and act as adversarial reconnaissance for an attacker. We design a unique threat model which leverages the LLM sycophancy effect and elevates the average attack success rate (ASR) from 17.7% to 86.2% in a multi-turn setting. We measure the mitigation effect of 7 black-box defense strategies, along with finetuning an open-source model to defend against leakage attempts.
  arXiv  Detail & Related papers  (2024-04-24T23:39:58Z)
• Exploring the Factual Consistency in Dialogue Comprehension of Large Language Models [51.75805497456226]
  This work focuses on the factual consistency issue with the help of the dialogue summarization task. Our evaluation shows that, on average, 26.8% of the summaries generated by LLMs contain factual inconsistency. To stimulate and enhance the dialogue comprehension ability of LLMs, we propose a fine-tuning paradigm with auto-constructed multi-task data.
  arXiv  Detail & Related papers  (2023-11-13T09:32:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.