Search papers, labs, and topics across Lattice.
Affiliation:
1
0
0
25
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.
