Search papers, labs, and topics across Lattice.
3
0
3
6
Existing security scanners misidentify nearly half of MCP server risks, challenging the reliability of current security assessments in LLM applications.
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.
