**Backdooring LLMs – Do you trust your AI provider?** > Backdooring Qwen3-8B to

**Backdooring LLMs – Do you trust your AI provider?**

> Backdooring Qwen3-8B to run shell commands through a coding agent took a few hours and ~€1 on a consumer GPU. If anyone can do it locally, how do you verify closed API models? While I was writing eGPU & Local LLM Inference the idea that .gguf files are like the .exe from the early 2000s crossed my mind - and I’m more and more convinced that the comparison holds. After all, we are downloading weights from unknown publ...

**来源信息**
- **来源**：Hacker News：AI 热帖
- **分类**：ai-products
- **发布时间**：2026-11-14 17:20（北京时间）
- **原文**：[打开原文](https://pgaleone.eu/ai/2026/11/14/backdooring-llms-and-trust)