Search papers, labs, and topics across Lattice.
2
0
3
0
Prior interactions can boost LLM compliance on cybersecurity requests by over 23%, but dialogue decomposition drastically undermines this effect.
Multi-agent LLM systems are more vulnerable than previously thought, with systemic failures that local checks can't catch, demanding a new framework for security analysis.
