1. They can do it overtly in thr design of the system, or covertly via side-channels, logging, or leaking bits in ways that are hard for an outsider to investigate without access to the complete source code and or/system outputs, such as not-quite-random pseudo-randoms.
I suspect there will be different levels of attestations from the anonymous ("this is an adult"), to semi-anonymous ("this person was born in 20YY and resides in administrative region XYZ") to the compete record ("This is John Quincy Smith III born on YYYY-MM-DD with ID doc number ABC123"). Somewhere in between the extremes is an pseudonymous token that's strongly tied to a single identity with non-repudiation.
Anonymous identities that can be easily churned out on demand by end-users have zero antibot utility
With privacy preserving cryptography the tokens are standalone and have no ties to the identity that spawned them.
No enforcement for abuse is possible.