I'd wager a guess that you set up some weird 'expression'? coupled with some bug in the IAM service, maybe some stale resources that you were deleting at the same time?
I'd then assume once expression is evaluated the services end up busy looping / proxies throwing out internal errors and taking out capacity.
Still, you shouldn't be able to cause downtime to more then a few servers in the extremely unlikely case I am anywhere close.
PS:
- I haven't used googles IAM so guessing after a few min of reading docs.
- you are incredibly unlikely to have triggered this at google's scale.