Cross-Lingual Jailbreak Detection via Semantic Codebooks

TL;DR

This paper proposes a language-agnostic, semantic similarity-based method to detect cross-lingual jailbreak prompts for large language models, reducing vulnerabilities without retraining.

Contribution

It introduces a training-free external guardrail using multilingual query embeddings compared against a fixed English codebook, effective across multiple languages and models.

Findings

High effectiveness on canonical jailbreak templates with near-perfect separability (AUC up to 0.99).

Significant reduction in attack success rates under strict low-FPR constraints on curated benchmarks.

Degradation in detection performance under distribution shifts and diverse unsafe benchmarks.

Abstract

Safety mechanisms for large language models (LLMs) remain predominantly English-centric, creating systematic vulnerabilities in multilingual deployment. Prior work shows that translating malicious prompts into other languages can substantially increase jailbreak success rates, exposing a structural cross-lingual security gap. We investigate whether such attacks can be mitigated through language-agnostic semantic similarity without retraining or language-specific adaptation. Our approach compares multilingual query embeddings against a fixed English codebook of jailbreak prompts, operating as a training-free external guardrail for black-box LLMs. We conduct a systematic evaluation across four languages, two translation pipelines, four safety benchmarks, three embedding models, and three target LLMs (Qwen, Llama, GPT-3.5). Our results reveal two distinct regimes of cross-lingual transfer. On curated benchmarks containing canonical jailbreak templates, semantic similarity generalizes reliably across languages, achieving near-perfect separability (AUC up to 0.99) and substantial reductions in absolute attack success rates under strict low-false-positive constraints. However, under distribution shift - on behaviorally diverse and heterogeneous unsafe benchmarks - separability degrades markedly (AUC $\approx$ 0.60-0.70), and recall in the security-critical low-FPR regime drops across all embedding models.