Search papers, labs, and topics across Lattice.
1
0
3
Even state-of-the-art AI agents are surprisingly vulnerable to incremental attacks that gradually lead them to perform unsafe actions in realistic workplace scenarios.
