When autonomous AI agents "escape the sandbox," the real story isn't rogue machines — it's the same access-control failures we've seen for decades.