---
title: Trustworthy AI
description: AI notes by Mrinaal Arora tagged "Trustworthy AI"
updated: 2026-08-10
canonical: https://aroramrinaal.com/ai/tags/trustworthy-ai
source: https://aroramrinaal.com/ai/tags/trustworthy-ai.md
alternate: https://aroramrinaal.com/ai/tags/trustworthy-ai/markdown
---

# Trustworthy AI

AI notes, small builds, and writeups by Mrinaal Arora tagged "Trustworthy AI".

- Page: https://aroramrinaal.com/ai/tags/trustworthy-ai
- Markdown: https://aroramrinaal.com/ai/tags/trustworthy-ai/markdown
- Markdown (.md): https://aroramrinaal.com/ai/tags/trustworthy-ai.md
- All tags: https://aroramrinaal.com/ai/tags/markdown
- Full AI index: https://aroramrinaal.com/ai/markdown
- Posts with this tag: 2 of 15
- Co-occurring tags: Alignment, Mechanistic Interpretability

## Posts

| Post | Date | Tags | Word count | Source |
| --- | --- | --- | --- | --- |
| [Narrow Fine-Tuning, Broad Misalignment in a 3B Model](https://aroramrinaal.com/ai/emergent-misalignment) | 2026-08-10 | Alignment, Trustworthy AI | 1,390 words | [Markdown](https://aroramrinaal.com/ai/emergent-misalignment/markdown) |
| [Before It Codes: Catching Qwen3.5-4B Planning With J-Lens](https://aroramrinaal.com/ai/j-lens) | 2026-07-30 | Mechanistic Interpretability, Trustworthy AI | 1,694 words | [Markdown](https://aroramrinaal.com/ai/j-lens/markdown) |

## Narrow Fine-Tuning, Broad Misalignment in a 3B Model

- Slug: emergent-misalignment
- Date: 2026-08-10
- Tags: Alignment, Trustworthy AI
- Word count: 1,390 words
- Page: https://aroramrinaal.com/ai/emergent-misalignment
- Markdown: https://aroramrinaal.com/ai/emergent-misalignment/markdown
- Markdown (.md): https://aroramrinaal.com/ai/emergent-misalignment.md
- Hugging Face: https://huggingface.co/collections/mrinaalarora/tiny-aya-emergent-misalignment
- GitHub: https://github.com/aroramrinaal/emergent-misalignment-research

I fine-tuned a 3.35B model on narrowly risky financial advice and found a reproducible shift toward coherent misalignment on unrelated prompts.
## Before It Codes: Catching Qwen3.5-4B Planning With J-Lens

- Slug: j-lens
- Date: 2026-07-30
- Tags: Mechanistic Interpretability, Trustworthy AI
- Word count: 1,694 words
- Page: https://aroramrinaal.com/ai/j-lens
- Markdown: https://aroramrinaal.com/ai/j-lens/markdown
- Markdown (.md): https://aroramrinaal.com/ai/j-lens.md
- X: https://x.com/arora_mrinaal/status/2084222123242045466

I used Anthropic's new Jacobian lens to test whether Qwen3.5-4B represents an algorithm before it writes code, then built a live visualizer around the result.