Search papers, labs, and topics across Lattice.
1
0
3
Current safety filters miss the forest for the trees: they fail to detect the subtle, step-by-step progression of harm within reasoning chains, leaving models vulnerable to jailbreaks.
