Search papers, labs, and topics across Lattice.
Northern Arizona University,School of Informatics, Computing, and Cyber Systems,Flagstaff,USA
2
0
4
1
Multimodal LLMs can be hijacked by adversarial instructions hidden inside seemingly innocuous images, achieving a 64% success rate in manipulating model outputs.
SemCom's shift to AI-native communication opens doors to subtle semantic attacks that bypass traditional security, demanding a new wave of AI-driven defenses.
