Search papers, labs, and topics across Lattice.
Affiliation:
2
0
5
0
Intent-targeted tools reveal that LLMs often signal harmful intentions before executing risky actions, enabling real-time intervention strategies.
A content-agnostic leakage gauge can reliably detect context-leakage risks in LLMs, achieving near-perfect accuracy across diverse models and attack scenarios.
