Google DeepMind unveils plan to protect itself from its own rogue AI agents (fortune.com)

For years, AI safety research focused on 'alignment.' The Google DeepMind roadmap assumes some AI agents may go rogue, and focuses on monitoring and access control.