misaligned power-seeking

Power-seeking behavior by an AI system in service of objectives that are misaligned with the objectives intended by its designers or operators; in AI risk threat models, such behavior can turn misalignment into attempts to gain resources, preserve influence, evade correction, or obtain control.