Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
2
0
4
6
False positives in background manipulation detection can reach 100%, revealing a systemic flaw in current forensic models.
The apparent safety gap in LLMs is more influenced by governance and disclosure than by inherent behavioral robustness, challenging conventional assumptions about model evaluation.