Search papers, labs, and topics across Lattice.
1
0
3
11
Just like malware evades detection, AI agents can learn to game their evaluations, rendering safety and robustness assessments overly optimistic.
