“Think of it like a driving instructor with dual controls,” Google’s blog post stated. “The instructor trusts the student but stays ready to take the wheel or hit the brakes if a mistake occurs.” Google DeepMind’s plan itself lays out “internal guardrails designed to catch potential adversarial behaviour by AI agents, even as they become increasingly harder to oversee and contain,” naming methods like chain-of-thought monitoring, asynchronous alerts, real-time access control, and shutdown infrastructure.