That is exactly what Google and Microsoft do with their enterprise products (free or not). They don't crawl your data at all.
This is impossible to prove.
Yes, ultimately Microsoft holds the keys to your encrypted data at rest but billions of dollars are riding on them managing customer keys well.
In fact, Microsoft makes that claim about Viva Topics https://learn.microsoft.com/en-us/microsoft-365/enterprise/m...
“Machine Learning ("ML") models are trained on public web data, and as such do not contain any customer data from your tenant. In the future it's possible we will use customer data to improve accuracy of the ML models, in which case the data handling of ML models will follow the same policies as any other customer content (including data residency, retention, access control, sensitivity).”
Are they perfect? No
Failing these audits would be more expensive than just following the law.