Let Execution Do the Teachingwith Sandeep Srinivasa, Arko Chattopadhyay, Enrique Martinho Ferrao, Guda Pranav Reddy1 versionW_97nm8f77v1 · current
Faithful Latent Harm Monitors Work No Better Than the Models Own Behaviour at a Matched False Alarm RateW_khby85hav1
Measuring Reward Hacking and Reasoning-Answer Decoupling Under Position-Confounded OptimizationSuyash Maniyar, Armaan Sandhu, Abhishek Mishra0 citations2608.15445v12026arXiv
Assessing Domain-Level Susceptibility to Emergent Misalignment from Narrow FinetuningAbhishek Mishra, Mugilan Arulvanan, Reshma Ashok et al.0 citations2602.00298v12026arXiv
Exploring Re-inforcement Learning via Human Feedback under User HeterogeneitySarvesh Shashidhar, Abhishek Mishra, Madhav Kotecha0 citations2601.20760v12026arXiv