What does that mean exactly?
Do you manually assess what is risky for a particular API, or is it up to the system to choose?
If it's up to it, what happens if it thinks that's not risky to delete user data?
You can also manually configure an allowlist/blocklist of operations for specific use cases.