Search papers, labs, and topics across Lattice.
1
0
3
LLM agents' internal activations leak detectable signals of collusion, even when transferred to structurally different multi-agent scenarios, opening a new frontier for AI safety.
