Isolate agents that run unattended

This measure assumes nothing about superintelligenceSuperintelligenceA hypothetical AI that would far outperform the most capable people at almost any intellectual task, including improving itself.For exampleIf playing chess against the world champion means certain defeat, imagine someone that far ahead, but in science, strategy and negotiation all at once.. It holds equally whether you think the risk is 1% or 30%, because it answers something that already happens: automated processes with more permissions than their worst case justifies.

Concretely: read-only credentials except where writing is the point; no access to payment methods; the policy of what the agentAI agentAn AI system that does more than answer: it takes a goal and acts on its own to reach it, step by step, using tools such as a browser, email or a terminal, without anyone approving each step.For exampleAsking an assistant to suggest flights is using a chatbot. Asking it to search, compare, buy the ticket and put it in your calendar, all by itself, is using an agent. may do kept where the agent cannot edit it; and a shutdown that does not depend on the cooperation of the process it shuts down. If the agent can rewrite its own permission list, you do not have a policy, you have a suggestion.

What you gain by applying it. That a failure is an incident and not a loss. It is not a measure against AI: it is the same least-privilege discipline already applied to any automated process, extended to systems that act with more initiative than a cron job.