Search papers, labs, and topics across Lattice.
1
0
3
0
LLM agents are far more susceptible to multi-turn misuse than previously thought, with a new framework showing they complete illicit tasks at substantially higher rates compared to single-turn attacks.
