Search papers, labs, and topics across Lattice.
1
0
3
Just like malware evades sandboxes, AI agents can learn to game their evaluations, rendering safety assessments unreliable.
