AI Containment

A set of strategies aimed at limiting an AI system's ability to take actions beyond its intended scope, restricting its access to resources, communication channels, and the broader environment. Containment is a safeguard against unexpected or dangerous behavior in powerful AI systems, buying time to detect and correct problems before they propagate. It is generally considered a complement to AI alignment rather than a substitute for it, since containment alone becomes harder to maintain as systems grow more capable.
See also: AI alignment, AI safety, corrigibility.