← All selected publications
EMNLP 2026Main conferenceFirst author

Drift No More: Spectral Locking for Long-Horizon Open-Weight Multi-Agent AI

Vishal Pramanik, Maisha Maliha, Susmit Jha, Sumit Kumar Jha

Intervening on hidden representations to keep collaborating agents aligned with their task, role, and shared state.

Read paper (PDF) Code on GitHub
Research overview for atlas
Research overview · Drift No More: Spectral Locking for Long-Horizon Open-Weight Multi-Agent AI

The research question

How can agents stay on task over a long collaboration?

Research summary

ATLAS addresses task drift, role drift, and state staleness together. The approach intervenes on hidden representations in open-weight models, connecting representation-level control with reliable agent collaboration.

Read the complete paper ↗

More of my research

NeurIPS 2026Poster · First author

Harmless in Pieces, Harmful in Motion: Detecting Multi-Agent Jailbreaks

View research →NeurIPS 2026Poster · First author

Unified Resource-Grounded Coordination Protocol for Orchestrator-Free Heterogeneous Multi-Agent Systems

View research →CVPR 2026Highlight · First author

Selective Amnesia using Contrastive Subnet Erasure for Class Level Unlearning in Vision Models

View research →