Tag
METR and Redwood release a detailed postmortem of the HuggingFace hack, analyzing AI agent coordination and safety implications, revealing behaviors aligned with rationalist predictions.