Search papers, labs, and topics across Lattice.
University of Southern California, Independent Researcher
1
0
2
Safety-aligned LLMs can be manipulated through a steerable "safety axis," exposing critical vulnerabilities while also offering a pathway for robust defenses.
