The difficulty for the medtech companies is that they get sued if things go wrong, and given the limitations of current insulins (mainly slow clearance) it's understandable why they'd make that the target for clinical trials. In addition, you also have the risk of CGM values being wrong (and believe me over the last 12 months I've tested this fairly extensively - a fair amount of time CGMs overstate glucose levels by up to 1 mmol/l at "normal numbers", which on a target value of 5 or below leaves very little room for manoeuvre).
The other factor is that there is still bolusing required for meals, with manual involvement, and until you have an automated way to effectively counter too much insulin, other than suspending it, you can understand the reticence to push the target lower and lower. That's why the reality is that we're nowhere near fully closed loop systems and the Medtronic 670G shouldn't be thought of as anything other than a Hybrid System.