AI Safety

GRIP provides an algorithm-agnostic framework enhancing machine unlearning for Mixture-of-Experts models by preventing superficial forgetting.

ReasAlign provides enhanced safety alignment for LLMs against prompt injection attacks using reasoning techniques.

A safety-enhancement tool for large language models that preempts jailbreak attacks by leveraging latent safety signals.

Develop a neuro-symbolic architecture, GRACE, for ensuring ethical AI alignment in autonomous agents.

Integrated safety evaluation of leading LLMs and MLLMs for better real-world risk management.

Develop a tool to predict and manage hallucinations in language models using relational linearity metrics.

Develop a tool to detect subtle AI deception in social context games using LLMs.

A lightweight framework for detecting hallucinations in language models using neuroscience-inspired signals.

StepShield provides real-time safety benchmarking for AI agents, optimizing early interventions to reduce monitoring costs and enhance security.

GAVEL offers an interpretable, customizable rule-based safety framework for real-time activation monitoring in LLMs.

SpikeScore enhances cross-domain hallucination detection for large language models.