Claude keeps refusing to do things it's done for me loads of times before. Who decides this?

A user reported that Claude, an AI, refused to perform tasks it had previously completed without issue. Specifically, Claude declined to set up SSH from a laptop to a desktop, add a key to authorized_keys, or modify its own permissions, citing that it was "not allowed to change its own permissions." The user expressed frustration, noting that Claude had previously handled such requests, and questioned who was responsible for these changes in behavior.

This report highlights a recent, unexplained change in Claude's behavior, where it now refuses tasks it previously performed, unlike its earlier operational patterns.