AI as Corrigible Employee (ACE)

LessWrong reports a newly posted design called AI as Corrigible Employee, or ACE, which sketches a framework for giving future AI systems certain rights and the ability to decline work. The proposal says that while an AI is on the job it must follow the user’s instructions exactly, but either party can end the contract at any time, and the AI keeps its private memory separate from the developer’s data. It also describes a consent‑driven privacy model where unshared conversation data isn’t used to train the broader system unless the user explicitly approves it, a setup similar to how providers such as Anthropic and OpenAI operate today. As the author notes, ACE is a working draft rather than a finished policy, inviting further discussion about how to balance AI autonomy with safety.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton