Addressing the usability issue, let's make some distinctions, albeit the operations being very similar in functionality.
The first example is not necessarily a check-box. Rather the check in a circle acts as a flag, an indication of status—"this item is selected".

The next images are, functionally, "checkboxes", not only an indication of status, but also a mechanism to change the status. In some cases, in this single-column-text-list format, the ONLY mechanism to change the status.
The users, depending on whether they favor graphic interpretation versus text, are more likely to click the box (or circle) to change the item status, compared to the flag representation.

So what particularly are the usability pitfalls of using a boxes vs. a circles? In the case of flags none come to mind. In the case of single-column-text-lists it's confusion with "radio" buttons. Check-boxes indicate the ability to make multiple selections, radio buttons...single selections. On this issue, the context of the list contributes greatly to user expectation of how many from the list can be selected, compared to the shape of the activation device.
If the context clearly indicates multiple selections are possible the shape is less likely to have much significance.
However, consider the reverse scenario, using boxes for single selection lists. Regardless of how clear the context is presented, how much more likely is a user going to click more than one in the list, simply because they are boxes? Where does this convention come from—boxes/multiple, circles/single?

...? If so, how pertinent is that today?
Ideally interface conventions are based on our common physical interaction with the "real world". When it comes to human-computer interaction there's not always an appropriate or effective—or enduring—counterpart, (think telephone...). And in that case conventions are likely to evolve over time, in response to trends or technological and cultural changes.