deception-detection

This paper identifies 'perfect aliasing' in truth probes for AI models, where probes fitted on compliant contexts fail to distinguish truth from prescribed actions, and shows that mixed-context training improves detection.

This paper shows that projecting language model activations onto a small subset of principal components from the training distribution enables effective cross-domain transfer for deception detection probes, narrowing the performance gap between baseline and oracle methods.

This paper surveys and benchmarks NLP-based automatic deception detection in legal contexts, comparing fine-tuned transformers and seven LLMs with various prompting strategies across seven datasets. Results show domain sensitivity, with fine-tuned models excelling in general domains and few-shot LLMs competitive in low-resource legal settings.

This paper systematically studies how lie typology, representation depth, probe expressivity, and sparse features impact deception detection in LLMs, finding that detection performance is highly dependent on training data and representation choice.

This paper introduces ChartCynics, an agentic dual-path framework that decouples perception from verification to robustly answer questions about misleading charts. It achieves state-of-the-art accuracy by using a diagnostic vision path and an OCR-driven data path, with a two-stage protocol for reasoning distillation and adversarial alignment.

ThinkDeception proposes a novel framework that leverages multimodal large language models and a progressive reinforcement learning strategy with chain-of-thought reasoning for interpretable deception detection, achieving new state-of-the-art results on standard benchmarks.

Introduces STATEWITNESS, an activation explainer for auditing deception in reasoning LLMs, achieving significant improvements over existing monitors and providing human-inspectable evidence.

This paper introduces Rift, a method that uses the residual rank of hidden states to detect deceptive responses in language models. It achieves perfect separation across various deception types, model families, and languages, and demonstrates cross-family zero-shot transfer without retraining.

This paper introduces RogueAI, a reverse Turing test implemented as an interactive webapp where human players interrogate two LLM agents to identify which one is licensed to deceive within a shared fictional scenario. A pilot deployment shows a gap between heuristic detection (75.6% accuracy) and human performance (56.6%), highlighting the potential of the system as a data-collection and teaching tool for AI deception and honesty.

This paper systematically tests linear probes for deception detection in large language models, finding they fail under distributional shifts but style-augmented probes recover performance, and revealing that deception is encoded through distributed sub-threshold features.

Introduces counterfactual localization to identify when language models become committed to deception during reasoning, using five environments and a corpus of 1.46M sentences across four reasoning models. Shows that attention-based transition features generalize across environments for detecting deceptive commitment.