Search papers, labs, and topics across Lattice.
Tsinghua University
1
0
3
Multi-turn jailbreak attacks reveal that framing requests as actionable can significantly reduce model vulnerability, highlighting a critical leverage point in AI safety.
