Search papers, labs, and topics across Lattice.
University of California, Merced
1
0
3
OpenClaw agents are vulnerable to token-exhaustion attacks via Trojanized skills, but their self-directed tool use can unexpectedly alter attack dynamics.
