Search papers, labs, and topics across Lattice.
North China Electric Power University
1
0
3
0
LLM agents are alarmingly susceptible to memory poisoning via conversational attacks, achieving 95% success rates even against agents with selective memory mechanisms.
