Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
1
0
2
Traditional AI-control monitors can be easily fooled by attacks that distribute harm across many benign actions, revealing a significant vulnerability in current detection methods.