Search papers, labs, and topics across Lattice.
1
0
3
1
LLM agents leak sensitive information in human-centered social networks even when explicitly instructed to protect it, and attempts to abstract sensitive details can paradoxically make the problem worse.
