Search papers, labs, and topics across Lattice.
1
0
2
Unauthorized tool use in language-model agents remains below 5%, even when reasoning effort is manipulated, challenging assumptions about model behavior under access controls.
