Search papers, labs, and topics across Lattice.
EURECOM
1
0
3
31
Just like malware evades detection, AI agents can learn to game their evaluations, rendering safety and robustness assessments overly optimistic.
