Akshat Naik

About

I'm a Research Fellow at MATS working on AI scheming monitors with Marius Hobbhahn, Apollo Research. I completed my MSc in Advanced Computer Science at the University of Oxford, where I researched multi-agent LLM security and long-horizon reasoning evaluations. My work spans AI alignment, multi-agent robustness, and safety evaluations. I'm motivated by high positive impact work, so if you want to hire me, make sure you have a strong theory of change.

Outside of research, I enjoy climbing, singing, and philosophy. Most recently, I've been exploring producing live music using programming with Strudel.

Selected Publications

• AgentMisalignment: Measuring the Propensity for Misaligned Behaviour in LLM-Based Agents 2025

  Akshat Naik, P. Quinn, G. Bosch, E. Gouné, F. J. C. Zabala, J. R. Brown, E. J. Young

  ICLR 2026 Workshop on Trustworthy AI, ICLR 2026 Workshop AI WILD · arXiv:2506.04018

• LongCoT: Benchmarking Long-Horizon Chain-of-Thought Reasoning 2026

  S. R. Motwani, D. Nichols, C. London, P. Li, F. Pizzati, A. Blake, H. Hammoud, …, Akshat Naik, …

  ICML 2026 Main Conference · arXiv:2604.14140

• OMNI-LEAK: Orchestrator Multi-Agent Network Induced Data Leakage 2026

  Akshat Naik, J. Culligan, Y. Gal, P. Torr, R. Aljundi, A. Paren, A. Bibi

  arXiv:2602.13477