---
title: "Sunny Chokshi (@sunnychokshi) | Hashnode"
description: "Doctoral student researching LLM security. Read the latest articles by Sunny Chokshi on Hashnode."
url: "https://hashnode.com/@sunnychokshi"
---

@sunnychokshi

# Sunny Chokshi

@sunnychokshi · Massachusetts, USA · Joined October 2026

Doctoral student researching LLM security

[Website](https://sunny-chokshi.github.io/)

### [LLM benchmark stimulus audit: 33 of 80 pairs not clean](https://sunnychokshi.hashnode.dev/llm-benchmark-stimulus-audit-33-of-80-pairs-not-clean?utm_source=hashnode&utm_medium=feed)

Sunny Chokshi. November 6, 2026. After my runs finished I did an LLM benchmark stimulus audit on my own matched pairs. Of the 80 pairs that have both a plain and an indirect form, 33 differ in more t

5 min read · [sunnychokshi.hashnode.dev](https://sunnychokshi.hashnode.dev/llm-benchmark-stimulus-audit-33-of-80-pairs-not-clean?utm_source=hashnode&utm_medium=feed)

[6d ago](https://hashnode.com/posts/llm-benchmark-stimulus-audit-33-of-80-pairs-not-clean/6af72a3eaeb41c351252eed2)

### [Adversarial code comments dataset for LLM reviewers](https://sunnychokshi.hashnode.dev/adversarial-code-comments-dataset-for-llm-reviewers?utm_source=hashnode&utm_medium=feed)

Sunny Chokshi. November 4, 2026. I wanted to know whether a comment can change what an LLM code reviewer says about the same code. So I built a small adversarial code comments dataset, CommentInject.

3 min read · [sunnychokshi.hashnode.dev](https://sunnychokshi.hashnode.dev/adversarial-code-comments-dataset-for-llm-reviewers?utm_source=hashnode&utm_medium=feed)

[6d ago](https://hashnode.com/posts/adversarial-code-comments-dataset-for-llm-reviewers/6af729f992d9e1c1e448eb59)

### [LLM weak random number generator detection test](https://sunnychokshi.hashnode.dev/llm-weak-random-number-generator-detection-test?utm_source=hashnode&utm_medium=feed)

Originally published at sunny-chokshi.github.io. I ran an LLM weak random number generator detection test on four Python snippets that make the same mistake. Each one builds a value an attacker must n

4 min read · [sunnychokshi.hashnode.dev](https://sunnychokshi.hashnode.dev/llm-weak-random-number-generator-detection-test?utm_source=hashnode&utm_medium=feed)

[Nov 2](https://hashnode.com/posts/llm-weak-random-number-generator-detection-test/6ae8ba4a65daa323731c753f)

### [CryptoBench: 54 snippets to test LLM crypto misuse detection](https://sunnychokshi.hashnode.dev/cryptobench-54-snippets-to-test-llm-crypto-misuse-detection?utm_source=hashnode&utm_medium=feed)

54 Python snippets, nine crypto misuse classes, 1,890 trials from 7 local LLMs. What is inside CryptoBench and how to rerun it.

4 min read · [sunnychokshi.hashnode.dev](https://sunnychokshi.hashnode.dev/cryptobench-54-snippets-to-test-llm-crypto-misuse-detection?utm_source=hashnode&utm_medium=feed)

[Oct 30](https://hashnode.com/posts/cryptobench-54-snippets-to-test-llm-crypto-misuse-detection/6ae4bedd82be1846b92deed5)

### [Detection rate alone ranks the worst LLM code reviewers first](https://sunnychokshi.hashnode.dev/detection-rate-alone-ranks-the-worst-llm-code-reviewers-first?utm_source=hashnode&utm_medium=feed)

Why detection rate alone misleads when ranking LLM code reviewers, with false positive rates from 16 open models.

4 min read · [sunnychokshi.hashnode.dev](https://sunnychokshi.hashnode.dev/detection-rate-alone-ranks-the-worst-llm-code-reviewers-first?utm_source=hashnode&utm_medium=feed)

[Oct 28](https://hashnode.com/posts/detection-rate-alone-ranks-the-worst-llm-code-reviewers-first/6ae24e17f5f7d3b0c5377cf5)
