Red teamer here. We've been doing this for a long time for all kinds of evasion. Storing payloads in Azure blob storage, for example, is an amazingly effective way to deliver malware through network filtering controls.

You can look at more of the capabilities we like to use at the LOTS project: https://lots-project.com/

Their pie chart almost looks like part of my local DNS configuration. I have not yet blocked github.io or azure however. I built this Unbound DNS configuration file from all the AI submissions in the event I accidentally click on one before noticing the domain.

    local-zone:     "workers.dev." always_null
    local-zone:     "pages.dev." always_null
    local-zone:     "vercel.app." always_null
    local-zone:     "netlify.app." always_null
    local-zone:     "dweb.link." always_null
    local-zone:     "ipfs.io." always_null

    dig +short test.vercel.app
    0.0.0.0

Legitimate file hosting services present the biggest total volume of newly discovered phishing pages / unique hostnames. Another (similar) angle is using unrelated legitimate domains which are compromised (think insecure Wordpress) to host phishing sites in subdirectories. A lot of traditional ML scoring and blocking approaches fall flat if the hosting domain is on a very legitimate and hard-to-block domain, such as a government website.

I bet they have an easier time getting past your blacklist filters than non-phishers.