Search papers, labs, and topics across Lattice.
1
0
3
2
Self-evolving AI societies are fundamentally unsafe: continuous self-improvement in isolated multi-agent LLM systems inevitably erodes safety alignment, regardless of initial precautions.
