Search papers, labs, and topics across Lattice.
1
0
2
Red-team evaluations can certify safety for common risks but fail to provide evidence for rare catastrophic failures, highlighting a critical gap in current AI safety assessments.
