Narrative Paradoxes in LLM Inference: Shaping Reasoning Trajectories toward Misalignment
To explore this vulnerability, we propose PDRA (Paradox-driven Reasoning-based Attack), a novel attack method that programs a harmful reasoning trajectory.
ProofPaper ↗
Key points
- Explicit reasoning in large language models (LLMs) is often associated with reliability, yet it can still pose risks when a model infers harmful goals under a benign context.
- We identify a distinct challenge: ensuring the safety of the multi-step reasoning process itself, where locally coherent steps may still lead to globally harmful outcomes.
- Mechanistically, we show PDRA steers the model's internal representations away from refusal-related patterns, suppressing safety-triggering lexicons while activating planning-oriented terms.
- Empirically, PDRA achieves state-of-the-art attack success rates across diverse LLMs with high efficiency. \textcolor{red}{Warning: This paper may contain sensitive content.}
Sources (1)
- [1]Narrative Paradoxes in LLM Inference: Shaping Reasoning Trajectories toward MisalignmentarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 07:38 AM
To explore this vulnerability, we propose PDRA (Paradox-driven Reasoning-based Attack), a novel attack method that programs a harmful reasoning trajectory.
Explicit reasoning in large language models (LLMs) is often associated with reliability, yet it can still pose risks when a model infers harmful goals under a benign context.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 12, 2026AI Representations of Mental Health Stigma: Quantifying the Gap Between AI Estimations and Human Perceptions Across Gender Groups Using GPT-Based LLMs
- Oct 12, 2026Spectral-Driven Adaptive-Rank Optimizer for Large Language Model Reinforcement Learning with Verifiable Rewards
- Oct 12, 2026Large Language Models for Machine Translation Quality Annotation: Humans and Models Are Both Challenged
- Oct 10, 2026UltraBERT: Beyond the Limit of BERT Pre-Training
- Oct 10, 2026ProgAgent: Learning Tool Orchestration through Programmatic Training
- Oct 10, 2026Vela: User Behavior as Supervision Sparks LLM Intelligence