Paper AI Control: Improving Safety Despite Intentional Subversion Greenblatt et al. · 2024 Read the paper → controlalignment
