Having worked in fraud detection, this is true for every provider you use. The adversarial nature of fraud detection means making _your definition of fraud_ public is not an option because people will use it as a loss function for their fraud.
I would just make sure I don't rely on a single provider and minimize single points of failure.
In reality bad guys operate at even larger scales than this, doing all kinds of things. Depending on what company you are, you might not even be able to pull the compute resources away from them right away; so in addition to burning disk alloc they're now burning compute units. It depends on the product too -- what if they spin up pubsub instances and leave messages in the queue? You might have to freeze those queues so that if they're valid customers they don't lose data. The more offerings you have, the more potential loss.
But yeah, they don't delete accounts right away -- there is indeed a grace period, and they do keep all your stuff (I suspect legislation may have something to do with it too). Last I remember it goes to appeal process (human review), and your data can't be deleted / resources reclaimed for quite a while. So it's not zero recourse for the user. Sounds like they're adding even more humans into the loop to ensure they don't unfortunately flag and then delete valid users. But it's non-trivial.
And what more important: there is plenty of services that will do lot's of check in advance and ask your passport / card / face photo and such, but they never ever stop your servers or destoy your data unless you do something extremely bad. And if it's come to company account it's even less likely to get in any trougle since most of services take work with companies serious.
Yet Google just like everyone around have extremely easy sign up process that all criminals are happy about, but can blow you in a face at any time. So they first create a problem to make marketing easier and then try to compensate for it by increasing risk of fault for thrir legit customers.
That's a good point. What's interesting is that Google's capabilities mean that they are able to allocate you the resources you ask for post-signup pretty much instantly, whereas AWS doesn't (they actually are just unable to allocate resources so quickly). This turns out to be a nice property if that n day spinup time window is enough to detect some bad guys and take them down in the interim, before they've caused massive damage to your infra / cost to your org. It's an O(1) improvement though. Bad guys will still just wait you out and then cause damage, so it's not like you're actually less likely to have your account frozen at AWS/Azure vs Google.
If you don't freeze the _bandwidth and compute_ resources you're even more fucked, because those have far worse scaling characteristics. So big company X taking immediate action to take down services make a lot of sense in that light. They don't do it in every situation though e.g borderline cases that require human review.
Fraud fighting is costly.
what i want to hear is "we can't make this mistake again".
This is basically the problem. You could say "this will never happen again", but instead it's "we don't want this to happen" as if it's somehow out of GCP's control.
As long as you will still delete accounts that you suspect of fraud, based on some opaque definition of suspecting fraud, "we don't want this to happen" is the best you can do, and that's not very reassuring.