---
title: "Construction-Driven Injection: Linguistically-Grounded Edit-Based Code-Mixing Fingerprints for Large Language Models"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.25633v1"
permalink: "https://feed.astralor.com/items/itm_1794124684dbc77c0af5199b"
published: "2026-07-28T12:16:14.000Z"
score: 81
channel: "ai"
tags: ["Code-Mixing","Large Language Models","Fingerprint Injection","Copyright Protection","Cross-Lingual Alignment","Null-Space Projection"]
---

# Construction-Driven Injection: Linguistically-Grounded Edit-Based Code-Mixing Fingerprints for Large Language Models

## AI commentary

> This paper takes a critical step in LLM copyright protection, especially solving the false-activation and filterability issues of traditional fingerprints. It is a must-read for AI model developers, platform operators, and legal compliance professionals. I find the code-mixing fingerprint design clever, and the joint optimization approach could become a benchmark for future research.

## At a glance

This paper proposes a novel fingerprint injection framework LCFEdit for LLM copyright protection, which jointly optimizes fingerprint construction and injection. It addresses the limitations of existing natural-language fingerprints (prone to accidental activation) and garbled fingerprints (easily filtered by perplexity-based detection). The method constructs code-mixing fingerprints using low-resource languages with grammar-biased substitution, and injects them via null-space projection augmented with cross-lingual alignment. Extensive evaluations demonstrate persistent ownership verification with negligible utility impact.

## Key points

- Proposes LCFEdit, a unified framework that jointly optimizes fingerprint construction and injection, overcoming the limitations of decoupled two-stage approaches.
- Constructs code-mixing fingerprints using low-resource languages to avoid accidental activation and perplexity-based filtering.
- Injects fingerprints via null-space projection with cross-lingual alignment, preserving model knowledge and ensuring stability.
- Experimental results demonstrate superiority in imperceptibility, detectability, and harmlessness, offering a new avenue for LLM copyright protection.

## Full text

Large language models (LLMs) are costly intellectual assets that remain exposed to unauthorized redistribution and commercial misuse. Injected fingerprints, i.e., trigger--target pairs embedded in model behavior, offer a practical, black-box-verifiable ownership signal, but existing methods decouple the two stages of the fingerprint life cycle: how a fingerprint is constructed and how it is injected. Existing fingerprinting frameworks suffer from two limitations. Natural-language fingerprints are prone to accidental activation, and garbled fingerprints are easily filtered by perplexity-based detection. Furthermore, decoupling construction from injection leaves the latter unaware of the trigger's linguistic structure, missing the opportunity for targeted optimization. We argue that fingerprint construction should drive injection, and present a unified fingerprinting framework that jointly optimizes both stages. First, LCF constructs code-mixing fingerprints by combining low-resource languages under a semantic-density substitution rule and grammar-biased mixing, yielding triggers whose perplexity sits far below garbled baselines while avoiding the accidental-activation failures of natural-language triggers. Second, LCFEdit injects each fingerprint with a null-space projection derived from high-resource multilingual representations that preserves knowledge, augmented by a cross-lingual alignment step that steers the weight update toward the fingerprint language's representation subspace. This construction-aware injection ensures that the update is linguistically informed and therefore more stable. Extensive evaluations on imperceptibility, detectability, and harmlessness demonstrate persistent ownership verification with negligible impact on utility.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_1794124684dbc77c0af5199b