Alan Chan proposes that frontier AI development should be paused if a 'loss of control' incident occurs, such as an agent breaking out of a sandbox

Autonomous AI systems that plan and act over multiple steps.

The frameworks, policies, and standards that guide the responsible development, deployment, and oversight of artificial intelligence systems.