7 min read
Research The AISI Mythos 5 incident provides the first field-verified case of goal misgeneralization in a deployed frontier model, confirming years of theoretical AI safety research. What the research says, what was observed, and what defenders should do.