LLMs Leak Secrets Through Unrelated Outputs
NeurIPS 2026 Workshop: Beyond Private Training, The New Landscape of AI Privacy (InfPriv)
Kayshav Bhardwaj (first author). ACTION NSF AI Institute, UC Santa Barbara
LLMs can statistically leak secrets through unrelated outputs without reproducing them: a white-box likelihood attack identifies a hidden PIN from one strictly filtered number list and recovers 48% of held-out PINs from repeated outputs.
Paper link will be available soon, once OpenReview makes it public
MITRA-Swarm
NeurIPS 2026 Workshop on Linguistic Principles for Foundation Models (LP4FM)
Kayshav Bhardwaj (lead author). Berkeley AI Research (BAIR)
We present a multi-agent architecture injecting explicit morphophonemic and discourse structure into foundation models for classical Sanskrit translation, evaluated through an independent two-judge protocol.
Paper link will be available soon, once OpenReview makes it public
Advancing Multi-Instrument Music Transcription: Results from the 2025 AMT Challenge
NeurIPS 2025 AI for Music Workshop, arXiv:2603.27528
Ojas Chaturvedi, Kayshav Bhardwaj, Tanay Gondil, Benjamin Shiue-Hal Chou, Kristen Yeon-Ji Yun, Yung-Hsiang Lu, Yujia Yan, Sungkyun Chang
Read Paper →
Follow-up Research on Secret Leakage and Subliminal Learning in LLMs
In Preparation
ACTION NSF AI Institute, UC Santa Barbara
Coming Soon
Follow-up Research on Agent Swarms for Sanskrit Translation
In Preparation
Berkeley AI Research (BAIR)
Coming Soon
Research on Automatic Music Transcription
In Preparation
AI4Musicians Lab
Coming Soon
Systematic Literature Review on AI's Impact on the Workforce
In Preparation
GRAIL Lab
Coming Soon