Omkar Neogi

Affiliation:

Papers on Lattice

Total citations

Topics

h-index

Research focus

Open-Source Models & Weights (1)Red-Teaming & Adversarial Robustness (1)Scaling Laws & Emergent Abilities (1)

Frequent co-authors

Alizishaan Khatri (1)C. Prabhu (1)

Papers (1)

Jun 22, 2026

Wrynx IncJun 22, 2026·also Affiliation:

Safety Beyond the Interface: Detecting Harm via Latent States in Large Language Models

This work extracts activations from LLaMA-3.1-8B and trains lightweight classifier probes (12.6M parameters) to detect harmful prompts and asks: does the model already know when the content is harmful?

Alizishaan Khatri, C. Prabhu, Omkar Neogi

Open-Source Models & Weights Red-Teaming & Adversarial Robustness Scaling Laws & Emergent Abilities

Omkar Neogi

Research focus

Frequent co-authors

Papers (1)

Search

Omkar Neogi

Research focus

Frequent co-authors

Papers (1)