It's both (over-automation and secondary sanctions).
The secondary sanctions exist because it's near-trivial to automatically create dozens of proxy accounts for a bad actor to launder reputation through. So in addition to direct fraud detection, Google has had to automate secondary "Is this account probably the same bad actor" detection.
That system, like any such system (including human review), has false-positivies where two accounts are believed to be the same owner when they are not. But the automation-with-insufficient-human-review problem is specifically Google's mistake, and there's room for improvement.
As somebody whose worked in the large scale anti-fraud area… I always make sure my team knows that the real people who fall through the cracks matter big time.
Anti-fraud is a tough space because you can never be 100% sure which actors are legit and which are scammers… after all if you knew who the scammers actually were you’d have blocked them all ready.
I always make sure there is some kind of escape hatch for legit users who get caught in our system. These escape hatches might not be super awesome for real users, especially if they fell into a bucket that strongly labels them as scammers, but at least they have an out.
The secondary sanctions exist because it's near-trivial to automatically create dozens of proxy accounts for a bad actor to launder reputation through. So in addition to direct fraud detection, Google has had to automate secondary "Is this account probably the same bad actor" detection.
That system, like any such system (including human review), has false-positivies where two accounts are believed to be the same owner when they are not. But the automation-with-insufficient-human-review problem is specifically Google's mistake, and there's room for improvement.