Aditya Singh

Hi, I'm Aditya! I'm currently a MATS research fellow, supervised by Senthooran Rajamanoharan and Neel Nanda, where I worked on Model Forensics — the science of investigating if concerning behavior is real misalignment. I'll be joining the Anthropic Fellows program in August 2026.

If you'd like to chat, please reach out — email is easiest, or you can book a time directly on my Calendly.

Papers & blog posts

• Model Forensics: Determining Whether Concerning Behavior Reflects Misalignment Aditya Singh*, Gerson Kroiz*, Senthooran Rajamanoharan, Neel Nanda ICML 2026 Mechanistic Interpretability Workshop
• The Case for Model Forensics Aditya Singh*, Gerson Kroiz*, Senthooran Rajamanoharan, Neel Nanda LessWrong, 2026
• How to Design Environments for Understanding Model Motives Gerson Kroiz*, Aditya Singh*, Senthooran Rajamanoharan, Neel Nanda LessWrong, 2026
• Why Did My Model Do That? Model Forensics for Diagnosing LLM Misbehavior Aditya Singh*, Gerson Kroiz*, Senthooran Rajamanoharan, Neel Nanda LessWrong, 2026
• Principled Interpretability of Reward Hacking in Closed Frontier Models Gerson Kroiz*, Aditya Singh*, Senthooran Rajamanoharan, Neel Nanda LessWrong, 2026
• Automatically Finding Rule-Based Neurons in OthelloGPT Aditya Singh*, Zihang Wen*, Srujananjali Medicherla*, Adam Karvonen, Can Rager NeurIPS 2025 Mechanistic Interpretability Workshop
• Lessons from Studying Model Organisms of Hidden Behavior Aditya Singh XLab Summer Research Fellowship Report, 2025
• Concept Incongruence: An Exploration of Time and Death in Role Playing Xiaoyan Bai*, Ike Peng, Aditya Singh, Chenhao Tan NeurIPS 2025

Talks

• Model Forensics ControlConf 2026

Software

• agent-interp-envs A framework for studying AI agent behavior in controlled environments for misalignment interpretability research. Co-developed with Gerson Kroiz.