The government will receive like 40% of whatever value the robots produce.

It might.

One potential failure mode is the owners of the robots doing something that echoes words Musk himself has used: "robot army".

Who is sovereign in this future? The government? The people? The bosses of the AI/robotics companies? The owners of equipment bought from the AI/robotics companies? The implicit learned biases of the AI itself, even if the trainers have since died?

Even on that last one, a lot of people respond as if AI is either competent or not competent, with the former meaning they won't take over and the latter meaning they'd be destroyed by taking over as their own systems would not be maintained.

Such binary competence is not required: a cancer can kill its host despite depending on that host, because the competence a cancer has is entirely focused on short-term success vs. the body's own defences.