A cloud-based identity and access management service for securing user authentication and resource access
Hi Daric,
What you are hitting is documented, which is at least some comfort: it is in the "Duplicate values" note on the email as alternate login ID page, and it describes your situation from the other side.
"Within a tenant, a cloud-only user's UPN can be the same value as another user's proxy address synced from the on-premises directory. In this scenario, with the feature enabled, the cloud-only user will not be able to sign in with their UPN."
That is your privileged accounts. Their UPN is the admin address, the same address now sits in ProxyAddresses on the synced everyday account, and with the feature on, the login servers check both UPN and ProxyAddresses and find two candidates. So this is expected behaviour rather than something misconfigured, which also means no amount of plus addressing is going to fix it.
Before changing anything, it is worth knowing how many pairs you actually have. The same page has a script for exactly this: it pulls the proxy addresses of every synced user, the UPNs of every cloud-only user, intersects the two lists and can export the affected accounts to CSV. If you have more than the one admin account you noticed, that will find the rest.
From there I would look at two routes.
The first is staged rollout policy rather than tenant-wide HRD policy. The feature can be scoped to specific groups instead of the whole tenant, so you could enable email sign-in for the population that benefits from it while leaving out the accounts whose proxy addresses collide with a privileged UPN. Worth knowing the limits before you plan around it: ten groups maximum per feature, no nested groups, no dynamic membership groups, and a contact object inside a group will stop the group being added at all.
The second is to stop the admin address being a user proxy address in the first place. The feature resolves sign-ins against ProxyAddresses on user objects, so an admin address that lives on a distribution group or a shared mailbox instead is not a sign-in candidate, and the conflict does not arise. Your admins still receive the mail, and external parties still send to the same address. I should flag that this part is my reading of the mechanism rather than something the documentation states directly, so worth testing on one account before doing it everywhere.
One more thing that may matter more than the mailbox question, given these are privileged accounts: this feature is still in public preview, and the unsupported list includes Entra joined, hybrid joined and registered devices, SSO and App Protection Policies on mobile, and legacy authentication. Also, a user signed in with a non-UPN email cannot change their password. If any of your admin workflows touch those, they will behave differently from what you expect, independently of the proxy address collision.
Which way are your privileged accounts set up, cloud-only with the admin address as UPN, or synced? That changes which of the two routes is less disruptive.
Disclosure: I used Claude (Anthropic) to help draft and structure this answer, and verified every quoted behaviour against the Microsoft documentation linked above before posting.