Downgrade untrusted label definitions

[mary-ext]

Is your feature request related to a problem? Please describe.

Moderation services have the ability to create new labels that are set to hide content by default.

This poses a serious risk as a new user testing out an (untrusted) moderation service could see parts of their feed hidden away from them without notice, and they'd have no idea until they noticed the label in question.

Describe the solution you'd like

Hide-by-default label definitions should be downgraded to warn-by-default.

There should be a toggle to "trust this moderation service" where it disables that downgrade functionality.

Describe alternatives you've considered

I would've just removed hide-by-default entirely but given that it mainly exists for Bluesky's own moderation service it's not really an option I suppose.

Additional context

N/A

[bnewbold]

It feels to me like the baseline is having apps/clients draw attention to "hide" defaults when signing up. Adding friction to labelers which intend to hide content (eg, abusive/harmful content) feels like unwanted friction to me.

A biggest concern to me, one that we discussed but I don't think ended up implementing yet, is preventing labelers from updating their declaration records in a way that expands their impact. Eg, a "badge-only" labeler could add a new label with default-hide behavior, and start hiding content. In that situation users wouldn't even realize they had granted permissions. I think that kind of update specifically should be treated like a mobile app requesting a change of permissions: get a notification/alert and need to approve (with option to un-subscribe or change settings) before the changes take effect.

[mary-ext]

This is sort of why I'm proposing an all-or-nothing preference where you can trust the labeler and allow their defaults.

It would be nice if we have diffs where you can review over the changes for sure, but I'm wary this could also be a little overwhelming? depends on how often labelers change their definitions.

I don't know how Bluesky app is implementing preferences for this but it's also not clear whether the preference for a specific label is the "default" or user-set.