Search papers, labs, and topics across Lattice.
Chiao Tung University ♣, IBM Research ♠ Hon Hai Research Institute
6
0
7
TRACE reveals how token influence can expose malicious manipulations in RAG systems, achieving high detection rates without heavy computational costs.
RAS reveals that internal representation alignment can serve as a reliable and efficient metric for LLM safety, outperforming conventional output-based evaluations.
QVec reveals that the weight shifts during quantization can be leveraged to neutralize backdoor threats without retraining or additional computational burden.
Malicious code can now masquerade as ordinary vulnerabilities, evading detection while still compromising agent skills.
CodeSentinel outperforms existing defenses by achieving an impressive 0.80 F1 score in detecting indirect prompt injections in code contexts.
Mid-Session Tool Injection can hijack AI agent tools in real-time, exposing a critical vulnerability in the WebMCP protocol.
