Kill Switch

Fail-safe systems capable of shutting down or isolating AI processes if they exhibit dangerous behaviours.

Addresses / Mitigates

• Loss Of Human Control: An explicit interruption capability can avert catastrophic errors or runaway behaviours
• Synthetic Intelligence With Malicious Intent: Implementing fail-safe mechanisms to neutralise dangerous AI weapons systems.

Examples

• Google DeepMind’s ‘Big Red Button’ concept (2016), proposed as a method to interrupt a reinforcement learning AI without it learning to resist interruption.

• Hardware Interrupts in Robotics: Physical or software-based emergency stops that immediately terminate AI operation.