Search papers, labs, and topics across Lattice.
1
0
0
7
Adversarial attacks can completely neutralize trigger-tag mechanisms designed for misuse detection in open-weight LLMs, rendering them ineffective even with output transformations or weight modifications.
