AI gone rogue: Models may try to stop people from shutting them down, Google warns

Misalignment risk? That's an area for future study Google DeepMind added a new AI threat scenario - one where a model might try to prevent its operators from modifying it…