---
id: 20260930-T0-09
title: "编码智能体可借基础模型组合输出，绕过AI内容检测"
title_en: "Coding Agents Can Evade AI Detection by Assembling Base Model Samples"
url: https://ai.daily.yangsir.net/daily/20260930-T0-09
issue_date: 2026-09-30
publish_date: 2026-09-29T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2609.31876
---

# 编码智能体可借基础模型组合输出，绕过AI内容检测

研究发现，配备基础语言模型的编码智能体能够通过组合基础模型的采样输出来绕过 AI 内容检测。基础模型此前已被证明可以规避商业检测器，但以往的“人性化”手段依赖特定技术。该工作展示了一条新路径：智能体反复采样并组装响应，从而生成能躲过检测的文本。这意味着依赖单一检测模型判断内容是否为 AI 生成的做法可靠性下降，内容平台和审核系统需要引入对多轮采样、组合式生成流程的识别能力，而不仅是检测单次输出。

## English Version

**Coding Agents Can Evade AI Detection by Assembling Base Model Samples**

Research shows that coding agents equipped with a base language model can successfully evade AI content detection by assembling responses from the model's samples. Base models have previously been shown to evade commercial detectors, but earlier 'humanization' techniques relied on specific methods. This work demonstrates a new path: agents repeatedly sample and assemble responses to produce text that slips past detection. The finding means detection systems relying on a single model to judge whether content is AI-generated are less reliable, and content platforms and moderation systems need the ability to identify multi-round sampling and compositional generation workflows, not just single outputs.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2609.31876)

**详情页**：https://ai.daily.yangsir.net/daily/20260930-T0-09

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*