Anonymizing / obfuscating mentions is modifying text that other people (not the person asking for the account action) wrote. That doesn’t make it necessarily wrong, but it is definitely a different decision and shouldn’t be made automatically, without explicit consent and intent.
By way of example, whether it is appropriate might depend on the particular terms and conditions of an individual instance, and what license and/or ownership grant each end user makes for what they write in that Discourse instance. If the T’s and C’s for a particular instance do not grant to the administrator the right to arbitrarily modify the “content” authored by site visitors, then automatically obfuscating mentions may inadvertently exceed the administrator’s agreed scope of action.
Additionally, this is a long-running “expensive” action that may adversely affect performance on a system with a lot of text content to scan and change. It may be operationally contraindicated on at least some sites, irrespective of license concerns, with respect to normal operations.