---
title: "Discussion on \"Reflections On Trusting Trust — Of Our AI Models\" | Hashnode"
description: "Discussion on \"Reflections On Trusting Trust — Of Our AI Models\". What \"local AI\" misses — could that \"open source\" model be backdoored? 🧐"
url: "https://hashnode.com/posts/trusting-trust-ai/6a50f6f7a25f4bc6553c51a4"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Reflections On Trusting Trust — Of Our AI Models](https://kevinhwang.dev/trusting-trust-ai?utm_source=hashnode&utm_medium=feed)

What "local AI" misses — could that "open source" model be backdoored? 🧐

Jul 10 · 10 min read · kevinhwang.dev  
Updated Oct 26  
Read the article

-   [#artificial-intelligence](https://hashnode.com/tag/artificial-intelligence)
-   [#llm](https://hashnode.com/tag/llm)
-   [#open-source](https://hashnode.com/tag/open-source)
-   [#security](https://hashnode.com/tag/security)
-   [#trust](https://hashnode.com/tag/trust)

## Written by

[Kevin Hwang](https://hashnode.com/@kevinhwang)

@kevinhwang

  
[View profile](https://hashnode.com/@kevinhwang)

Señor Vibe Coder

## Responses 1

Markdown supported · ⌘↵ to post

[Ahmet Özel](https://hashnode.com/@ahmetozel)

[Jul 12](https://hashnode.com/posts/trusting-trust-ai/6a50f6f7a25f4bc6553c51a4/comment/6a52fd64d4edccbbe2c63cbb)

Open weights improve inspectability, but they do not provide a reproducible supply chain by themselves. For production use I would want immutable weight and tokenizer digests, a signed model card, training/fine-tuning lineage, exact inference configuration, and an eval suite tied to the deployed artifact. Runtime controls still matter too: tool allowlists, retrieval-source policy, output validation, and traceable policy versions. Trust becomes measurable when the deployed binary, configuration, and evaluation evidence can all be reproduced.