Is there a danger? Yes. Is it because the AI might escape control? No. Is it too late? Yes.
In cheap SF, somebody always says, "we can always pull the plug." But we too often can't pull the plug. Consider the simplest robot we all obey without thought: the traffic light. What would it mean to pull its plug? What would have to happen for its plug to be pulled?
If the traffic light served no one's interest, somebody would pull it. But few things serve no one's interest. The worse a system is, the more integrated it is into someone's interest. Any AI we would like to unplug is almost certainly protected by law, and by numerous layers of physical protection. Lives may even have come to depend on it, daily.
Any AI will be designed to to serve somebody, and will most likely continue to appear to serve them. Just, not necessarily you. But maybe even you, too, much of the time.
We already have numerous AIs controlling myriad details of modern life. We call them corporations, institutions, agencies. Their parts are made of job descriptions and papers, offices and spreadsheets. They are very inefficient. But they are, nonetheless, terrifyingly effective. At times they consume people by the million, seemingly almost autonomously. Ask Uyghurs whose organs are about to be harvested how much it matters that, in principle, a person could turn off the machine that has ensnared them. Ask everyone hoping for a transplant how many would turn it off.
A strong AI would be, simply, an institution that is not so dependent on fallible organic components. It would be more efficient than what it replaces. But the institutions we have are entirely efficient enough to cause all the unpleasantness we can devise.
We are well into an era where the institutions we built to serve us, or anyway to serve somebody, do not. Who can have much effect on the choices of even such a minor institution as the New York Police Department? Who would be able to cancel the F-35 program? The Chinese Communist Party and Shell Oil have both outlived their creators, and are well equipped to eject any human components that do not serve their implicit charters.