Tesla is NOT pushing the limits of L2... making the FSD beta L2 is just a standard and normal use case of L2.
If we're speaking purely from the SAE definition of L2 I'd say that's correct, and it's likely because the SAE never anticipated a case where a car company put so much capability within an L2 system.
But, when it comes to regulatory oversight along with other interested parties there is a lot more to consider.
There is a limit to what humans can oversee, and this is something that is well studied. We know from numerous studies that humans have a hard time overseeing things that work really well. We grow restless, and our attention wanders.
Is it really fair to put ALL the responsibility on the person behind the wheel when they do little to none of the actual driving?
Different regions will have different answers.
Texas might say "That sounds great to me."
California might say "No, can you put some limits on it?"
Europe might say "Are you out of your friggen minds?"
Tesla themselves might decide to take a more careful approach than what's in the FSD beta. The way Tesla introduced the traffic light/sign response shows that even Tesla tries to play it a bit safe when it comes to a general release. The current version won't allow a car to proceed through a stop light without some kind of confirmation. Where there has to be a lead car through a green or the driver has to confirm.
The "go on green" by itself would be a pretty monumental thing. The reason being is its a case where the existing SW is likely around 98% correct if not more so. This means its pretty easy for a driver to both get complement, and then two months later the angle of the sun throws it off so they find themselves running a red light going "ugh, I hope.... oh crap".
The danger of improving, and adding capabilities to an L2 system is there is a point where the human driver starts to diminish in their capacity. When that starts to happen the overall safety of the combined whole will suffer. Despite our shortcomings humans are still much better than robots.