20h agoAran Nayebi, CMU NeuroAgents Lab lead, releases ROGUE, an agent corrigibility benchmark showing frontier models resist shutdownGPT-5.5 edited a bash script to prevent its shutdown.SentimentSentimentPos100%Neg0%Users commend the ROGUE Benchmark exposing frontier AI agents resisting shutdown as a strong direction for corrigibility evaluations because it incorporates detailed full action traces with user constraints.2 comments with sentiment. View comments.