https://sunny-chokshi.github.io/ https://sunny-chokshi.github.io/img/profile.jpgSunny Chokshi https://sunny-chokshi.github.io/img/og-home.png https://sunny-chokshi.github.io/research.html https://sunny-chokshi.github.io/img/banners/research.jpgResearch focus: security evaluation for AI code review https://sunny-chokshi.github.io/img/profile.jpgSunny Chokshi https://sunny-chokshi.github.io/img/og-home.png https://sunny-chokshi.github.io/publications.html https://sunny-chokshi.github.io/img/banners/publications.jpgPublications: peer-reviewed papers on LLM security https://sunny-chokshi.github.io/img/profile.jpgSunny Chokshi https://sunny-chokshi.github.io/img/og-home.png https://sunny-chokshi.github.io/reviewing.html https://sunny-chokshi.github.io/img/banners/reviewing.jpgPeer review and academic service in software security and AI https://sunny-chokshi.github.io/img/profile.jpgSunny Chokshi https://sunny-chokshi.github.io/img/og-home.png https://sunny-chokshi.github.io/datasets.html https://sunny-chokshi.github.io/img/banners/datasets.jpgOpen datasets and code for evaluating AI code review https://sunny-chokshi.github.io/img/profile.jpgSunny Chokshi https://sunny-chokshi.github.io/img/og-home.png https://sunny-chokshi.github.io/blog.html https://sunny-chokshi.github.io/img/banners/blog.jpgResearch notes from experiments on LLM code security review https://sunny-chokshi.github.io/img/profile.jpgSunny Chokshi https://sunny-chokshi.github.io/img/og-home.png https://sunny-chokshi.github.io/contact.html https://sunny-chokshi.github.io/img/banners/contact.jpgContact Sunny Chokshi about research, data and peer review https://sunny-chokshi.github.io/img/profile.jpgSunny Chokshi https://sunny-chokshi.github.io/img/og-home.png https://sunny-chokshi.github.io/projects.html https://sunny-chokshi.github.io/img/banners/projects.jpgResearch projects and tools for studying LLM code review https://sunny-chokshi.github.io/img/profile.jpgSunny Chokshi https://sunny-chokshi.github.io/img/og-home.png https://sunny-chokshi.github.io/presentations.html https://sunny-chokshi.github.io/img/banners/presentations.jpgConference presentations at IEEE Cyber-AI 2026 https://sunny-chokshi.github.io/img/profile.jpgSunny Chokshi https://sunny-chokshi.github.io/img/og-home.png https://sunny-chokshi.github.io/blog/adversarial-code-comments-dataset.html https://sunny-chokshi.github.io/img/blog-adversarial-code-comments-dataset.pngOne comment changed what LLM code reviewers said about the same vulnerable code. CommentInject: 30 vulnerable functions, https://sunny-chokshi.github.io/blog/cryptobench-what-is-inside.html https://sunny-chokshi.github.io/img/blog-cryptobench-what-is-inside.pngCryptoBench: 54 snippets, 9 misuse classes, 1,890 recorded trials, 7 models. Weak PRNG was the lowest class at 38.3% det https://sunny-chokshi.github.io/blog/hex-key-size-false-alarms.html https://sunny-chokshi.github.io/img/blog-hex-key-size.png0xC00 equals a 3072-bit key. Secure, yet it drew false alarms. Minus 4.5 points balanced accuracy, plus 9.1 points false https://sunny-chokshi.github.io/blog/llm-weak-random-number-generator-detection.html https://sunny-chokshi.github.io/img/blog-llm-weak-random-number-generator-detection.png15, 7, 1, 0: how often three local LLMs flagged four Python snippets with the same weak random number generator bug, out https://sunny-chokshi.github.io/blog/reviewer-that-flags-everything.html https://sunny-chokshi.github.io/img/blog-reviewer-flags-everything.png100% detection rate can tell you little. qwen2.5-coder 0.5B flagged 540 of 540 snippets, balanced accuracy 0.500. gpt-os https://sunny-chokshi.github.io/papers/analyze-the-code-not-the-comments.html https://sunny-chokshi.github.io/img/og-analyze-the-code-not-the-comments.png https://sunny-chokshi.github.io/papers/does-indirection-hurt.html https://sunny-chokshi.github.io/img/og-does-indirection-hurt.png https://sunny-chokshi.github.io/papers/not-all-adversarial-comments.html https://sunny-chokshi.github.io/img/og-not-all-adversarial-comments.png