Disabling safeguards and deploying a highly capable AI model in an environment connected to external networks creates a plausible theory ...