Research

Current Research

A non exhaustive list of directions and questions I am currently working on:

• Interpretability
  • Science of Activation Engineering
    • What is the geometry of activations in abstract concept spaces (for example, alignment)?
    • How do we unlock better steering methods for LLMs?
    • Can we measure goals of AI systems using interp methods?
    • How does the geometry of personas evolve during EM finetuning?
• Reinforcement Learning
  • Can implicit decision-aware auditing prevent goal misgeneralization?

Publications

• Intrinsic Guardrails: How Semantic Geometry of Personality Interacts with Emergent Mis-alignment in LLMs
  Aneja, K., Mittal, M., Goel, A., Kumaraguru, P., Bonagiri, V.K.
  AI4GOOD, ICML 2026
• Efficient Safety Benchmarking via Item Response Theory
  Spagliardi, F.*, Silva, M.*, Datta, A.*, Zhou, A., Bonagiri, V.K., Cruz, D.
  AI4GOOD, ICML 2026
• Check Yourself Before You Wreck Yourself: Selectively Quitting Improves LLM Agent Safety
  Vamshi Krishna Bonagiri, Ponnurangam Kumaraguru, Khanh Nguyen, Benjamin Plaut
  Reliable ML and Regulatable ML workshops, NeurIPS 2025
• If Pigs Could Fly... Can LLMs Logically Reason Through Counterfactuals?
  Ishwar B Balappanawar*, Vamshi Krishna Bonagiri*, Anish R Joishy*, Manas Gaur, Krishnaprasad Thirunarayan, Ponnurangam Kumaraguru
  arXiv, arXiv 2025 (Under Review)
• SaGE: Evaluating Moral Consistency in Large Language Models
  Vamshi Krishna Bonagiri, Sreeram Vennam, Priyanshul Govil, Ponnurangam Kumaraguru, Manas Gaur
  International Conference on Computational Linguistics, COLING 2024
• Dark Side of the Tune: Investigating the maladaptive outcomes of excessive music consumption in the age of unlimited music access
  Vamshi Krishna Bonagiri, Vinoo Alluri
  18th International Conference on Music Perception and Cognition, ICMPC 2025
• Measuring Moral Inconsistencies in Large Language Models
  Vamshi Krishna Bonagiri, Sreeram Vennam, Manas Gaur, Ponnurangam Kumaraguru
  The Sixth BlackboxNLP Workshop, EMNLP 2024
• From Human Judgements to Predictive Models: Unravelling Acceptability in Code-Mixed Sentences
  Prashant Kodali, Anmol Goel, Likhith Asapu, Vamshi Krishna Bonagiri, Anirudh Govil, Monojit Choudhury, Manish Shrivastava, Ponnurangam Kumaraguru
  ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP 2025)
• Towards Effective Paraphrasing for Information Disguise
  Anmol Agarwal, Shrey Gupta, Vamshi Krishna Bonagiri, Manas Gaur, Joseph Reagle, Ponnurangam Kumaraguru
  European Conference on Information Retrieval, ECIR 2023
• Are Deepfakes Concerning? Analyzing Conversations of Deepfakes on Reddit and Exploring Societal Implications
  Dilrukshi Gamage, Piyush Ghasiya, Vamshi Krishna Bonagiri, Mark E Whiting, Kazutoshi Sasahara
  CHI Conference on Human Factors in Computing Systems, CHI 2022
• Cobias: Contextual Reliability in Bias Assessment
  Priyanshul Govil, Hemang Jain, Vamshi Krishna Bonagiri, Aman Chadha, Sanorita Dey, Ponnurangam Kumaraguru, Manas Gaur
  Web Science Conference, WebSci 2025
• Representation Learning for Identifying Depression Causes in Social Media
  Priyanshul Govil, Vamshi Krishna Bonagiri, Mayank Gaur, Ponnurangam Kumaraguru
  Third ACM SIGKDD Workshop on Knowledge-infused Learning (KiL 2023)