# Path Drift in Large Reasoning Models:How First-Person Commitments Override Safety

arXiv: https://arxiv.org/abs/2510.10013 · HTML: https://mlsanity.com/paper/2510.10013/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning.
- Builds on 18 of its 26 references that are in the map.
- First posted on arXiv 11 Oct 2025.

## Placed in

- LLM reasoning / [Prompting and chain-of-thought reasoning](https://mlsanity.com/lr/prompting-and-chain-of-thought-reasoning/)

## Builds on (most cited first, 18 of 18)

- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools](https://mlsanity.com/paper/2406.12793/) (2024)
- [Kimi-VL Technical Report](https://mlsanity.com/paper/2504.07491/) (2025)
- [Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models](https://mlsanity.com/paper/2503.09567/) (2025)
- [Demystifying Long Chain-of-Thought Reasoning in LLMs](https://mlsanity.com/paper/2502.03373/) (2025)
- [ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learning](https://mlsanity.com/paper/2504.01296/) (2025)
- [When More is Less: Understanding Chain-of-Thought Length in LLMs](https://mlsanity.com/paper/2502.07266/) (2025)
- [SafeChain: Safety of Language Models with Long Chain-of-Thought Reasoning Capabilities](https://mlsanity.com/paper/2502.12025/) (2025)
- [H-CoT: Hijacking the Chain-of-Thought Safety Reasoning Mechanism to Jailbreak Large Reasoning Models, Including OpenAI o1/o3, DeepSeek-R1, and Gemini 2.0 Flash Thinking](https://mlsanity.com/paper/2502.12893/) (2025)
- [The Hidden Risks of Large Reasoning Models: A Safety Assessment of R1](https://mlsanity.com/paper/2502.12659/) (2025)
- [Reasoning-Augmented Conversation for Multi-Turn Jailbreak Attacks on Large Language Models](https://mlsanity.com/paper/2502.11054/) (2025)
- [Phi-4-Mini-Reasoning: Exploring the Limits of Small Reasoning Language Models in Math](https://mlsanity.com/paper/2504.21233/) (2025)
- [A Mousetrap: Fooling Large Reasoning Models for Jailbreak with Chain of Iterative Chaos](https://mlsanity.com/paper/2502.15806/) (2025)
- [Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillation](https://mlsanity.com/paper/2503.16385/) (2025)
- [Unlocking General Long Chain-of-Thought Reasoning Capabilities of Large Language Models via Representation Engineering](https://mlsanity.com/paper/2503.11314/) (2025)
- [SafeMLRM: Demystifying Safety in Multi-modal Large Reasoning Models](https://mlsanity.com/paper/2504.08813/) (2025)
- [Refining Input Guardrails: Enhancing LLM-as-a-Judge Efficiency Through Chain-of-Thought Fine-Tuning and Alignment](https://mlsanity.com/paper/2501.13080/) (2025)

## Built on by (most cited first, 5 of 5)

- [dVLM-AD: Enhance Diffusion Vision-Language-Model for Driving via Controllable Reasoning](https://mlsanity.com/paper/2512.04459/) (2025)
- [The Model Knows, the Decoder Finds: Future Value Guided Particle Power Sampling](https://mlsanity.com/paper/2605.02427/) (2026)
- [Multi2: Hierarchical Multi-Agent Decision-Making with LLM-Based Agents in Interactive Environments](https://mlsanity.com/paper/2606.03698/) (2026)
- [Visioning Human-Agentic AI Teaming: Continuity, Tension, and Future Research](https://mlsanity.com/paper/2603.04746/) (2026)
- [Reach Into The CHOIR: Free-List Elicitation Uncovers Distinct Model Voices in LLM Ensembles](https://mlsanity.com/paper/2609.38448/) (2026)

## Abstract (arXiv)

As large language models (LLMs) are increasingly deployed for complex reasoning tasks, Long Chain-of-Thought (Long-CoT) prompting has emerged as a key paradigm for structured inference. Despite early-stage safeguards enabled by alignment techniques such as RLHF, we identify a previously underexplored vulnerability: reasoning trajectories in Long-CoT models can drift from aligned paths, resulting in content that violates safety constraints. We term this phenomenon Path Drift. Through empirical analysis, we uncover three behavioral triggers of Path Drift: (1) first-person commitments that induce goal-driven reasoning that delays refusal signals; (2) ethical evaporation, where surface-level disclaimers bypass alignment checkpoints; (3) condition chain escalation, where layered cues progressively steer models toward unsafe completions. Building on these insights, we introduce a three-stage Path Drift Induction Framework comprising cognitive load amplification, self-role priming, and condition chain hijacking. Each stage independently reduces refusal rates, while their combination further compounds the effect. To mitigate these risks, we propose a path-level defense strategy incorporating role attribution correction and metacognitive reflection (reflective safety cues). Our findings highlight the need for trajectory-level alignment oversight in long-form reasoning beyond token-level alignment.