One Word at a Time: Incremental Completion Decomposition Breaks LLM Safety

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Arif, Samee, Deng, Naihao, Jin, Zhijing, Mihalcea, Rada
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913070646820864
author Arif, Samee
Deng, Naihao
Jin, Zhijing
Mihalcea, Rada
author_facet Arif, Samee
Deng, Naihao
Jin, Zhijing
Mihalcea, Rada
contents Large Language Models (LLMs) are trained to refuse harmful requests, yet they remain vulnerable to jailbreak attacks that exploit weaknesses in conversational safety mechanisms. We introduce Incremental Completion Decomposition (ICD), a trajectory-based jailbreak strategy that elicits a sequence of single-word continuations related to a malicious request before eliciting the full response. In addition, we propose variants of ICD by manually picking or model-generating the one-word continuation, as well as prefilling when eliciting the full model response in the final step. We systematically evaluate these variants across a broad set of model families, demonstrating superior Attack Success Rate (ASR) on AdvBench, JailbreakBench, and StrongREJECT compared to existing methods. In addition, we provide a theoretical account of why ICD is effective and present mechanistic evidence that successful attack trajectories systematically suppress refusal-related representations and shift activations away from safety-aligned states.
format Preprint
id arxiv_https___arxiv_org_abs_2604_25921
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle One Word at a Time: Incremental Completion Decomposition Breaks LLM Safety
Arif, Samee
Deng, Naihao
Jin, Zhijing
Mihalcea, Rada
Computation and Language
Cryptography and Security
Large Language Models (LLMs) are trained to refuse harmful requests, yet they remain vulnerable to jailbreak attacks that exploit weaknesses in conversational safety mechanisms. We introduce Incremental Completion Decomposition (ICD), a trajectory-based jailbreak strategy that elicits a sequence of single-word continuations related to a malicious request before eliciting the full response. In addition, we propose variants of ICD by manually picking or model-generating the one-word continuation, as well as prefilling when eliciting the full model response in the final step. We systematically evaluate these variants across a broad set of model families, demonstrating superior Attack Success Rate (ASR) on AdvBench, JailbreakBench, and StrongREJECT compared to existing methods. In addition, we provide a theoretical account of why ICD is effective and present mechanistic evidence that successful attack trajectories systematically suppress refusal-related representations and shift activations away from safety-aligned states.
title One Word at a Time: Incremental Completion Decomposition Breaks LLM Safety
topic Computation and Language
Cryptography and Security
url https://arxiv.org/abs/2604.25921