---
id: 20260729-T0-02
title: "Semalith v1.4：184M参数分类器，44倍参数少于Llama-Guard-3，实现SOTA提示注入检测"
title_en: "Semalith v1.4: 184M Safety Classifier Beats 8B Llama-Guard-3 in Prompt Injection Detection"
url: https://ai.daily.yangsir.net/daily/20260729-T0-02
issue_date: 2026-07-29
publish_date: 2026-07-28T04:00:00.000Z
category: release
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2607.22545
---

# Semalith v1.4：184M参数分类器，44倍参数少于Llama-Guard-3，实现SOTA提示注入检测

Semalith AI发布了v1.4版本的安全分类器Semalith，参数量仅184M，比Llama-Guard-3-8B少44倍，但在提示注入、监管合规和有害内容检测上达到SOTA水平。该模型专为金融服务和智能体环境设计，填补了现有开源护栏无法同时覆盖这三个维度的空白。开发者可直接部署用于过滤恶意输入。

## English Version

**Semalith v1.4: 184M Safety Classifier Beats 8B Llama-Guard-3 in Prompt Injection Detection**

Semalith AI released v1.4 of its safety classifier with 184M parameters, achieving state-of-the-art performance on prompt injection, regulatory compliance, and harm detection at 44x fewer parameters than Llama-Guard-3-8B. It targets financial and agentic AI deployments. Available for open-source use.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2607.22545)

**详情页**：https://ai.daily.yangsir.net/daily/20260729-T0-02

---

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