My best guesses:
1. The number of automated scams/attacks and associated support requests is unbounded vs. bounded human labor so it's a losing investment.
2. Machine learning is sufficient for attackers to undo the anti-abuse work on a low number of false positives from human intervention. Throw small behavioral variants of banned scam/attack accounts at support and optimize for highest reinstatement rate. This abuse traffic will be the bulk of what the humans have to deal with.
3. They'd probably be hiring a non-negligable percentage of the same people who are running scams. The risk of insider abuse is untenable.
This is the first time I hear someone making this claim. Is there prior evidence of this being a regular occurrence with outsourced customer support operations?
1. OP specifically said offshore hires presumably for cheaper wages. Anywhere wages are currently cheap there's a greater incentive to run Internet scams: it's farther from law enforcement agencies that care, alternate employment doesn't pay as well, there's even a culture of acceptability in some countries where trickling money from richer nations is seen as a net benefit to the local society.
2. Google is a high profile target. Scammers will try to get hired, existing workers will get bribed or realize the opportunity they have.
I don't have any scientific evidence. https://www.google.com/amp/s/news.abs-cbn.com/amp/business/0... is one instance of Google having to switch vendors for fraud in a non-1st-world country.