Search papers, labs, and topics across Lattice.
AIT Austrian Institute of Technology
4
0
6
17
Adversarial prompt injections can trick LLMs into misclassifying malicious log entries as benign, jeopardizing cybersecurity efforts.
Knowledge graphs can boost RAG performance by 35% on complex, multi-hop cyber threat intelligence queries, but only when combined with traditional text retrieval.
A new dataset of labeled system logs, CAM-LDS, enables training and evaluation of LLMs for automated, domain-agnostic cyberattack analysis.
Achieve adaptive, efficient intrusion detection in distributed settings without heavy computation by dynamically reconfiguring detectors based on available resources.
