Search papers, labs, and topics across Lattice.
1
0
3
1
Pre-load auditing of Agent Skills can achieve >97% accuracy in detecting malicious intent, even against semantics-preserving rewrites, by combining role-aware evidence extraction with semantic verification.
