Search papers, labs, and topics across Lattice.
6
0
5
8
Evidence-grounded detection in FlowGuard reveals that traditional semantic analysis can miss critical execution-related risks, achieving up to 2.23x faster scanning.
Existing security scanners misidentify nearly half of MCP server risks, challenging the reliability of current security assessments in LLM applications.
AEGIS achieves near-zero attack success rates against visual synonym jailbreaks while preserving the fidelity of benign outputs, reshaping the defense landscape for text-to-image models.
Current LLMs can autonomously penetrate systems with success rates up to 69.3%, revealing alarming implications for cybersecurity.
Prevent LLM agents from leaking data or causing financial loss with AgentGuard, a new access control framework that requires only ~10 lines of code to integrate.
A targeted tweak to positional embeddings can neutralize unsafe content generation in text-to-image transformers without sacrificing image quality.
