A Love Wins Media product

Safety on AudienceKit

Big-platform safety.
Your people decide.

Machines flag. People make the calls that matter.

The tools that keep the largest platforms safe usually need a trust and safety department to run. AudienceKit brings them to communities of every size: reports and automatic checks catch problems as they appear, an AI host helps in chat, and anything serious waits for someone from your own team.

That someone works in Arbiter, our moderation app for iPhone, iPad and Mac.

How a problem is handled

Flag, help, decide

Each layer hands what it can’t settle to the next, and the last one is always a person.

Automated tools flag

Member reports, blocks, automatic message checks and your own word lists catch problems the moment they appear, so nobody has to watch every room.

An AI host helps

Where an audience turns it on, Arbiter’s AI host reads venue chat as it happens, answers questions, handles the small stuff with a private nudge, and proposes what to do about the rest.

People decide

Anything the AI isn’t sure of, and anything punitive, waits for a person in Arbiter’s queue. Every action is logged with its reason, and any ban can be lifted.

Arbiter's gold balance scale with glowing glass pans

Arbiter · iPhone, iPad & Mac

The calls the AI
shouldn’t make alone.

Arbiter is where your administrators review what AudienceKit’s AI decided, and decide what it couldn’t.

AudienceKit’s AI handles routine work on its own: listing suggestions, research on submissions and moderation calls. Arbiter shows your team the rest, in three lists that stay live as people work:

  • Needs you. What the AI wasn’t sure of, couldn’t answer, or isn’t allowed to decide alone, the least certain first.
  • Decided by AI. What it was confident about and applied on its own, so you can check its work.
  • Decided by people. What your team decided, and whether they went the AI’s way.

Each decision shows the AI’s answer with a probability for every option, its reasoning and evidence, and the options it couldn’t rule out. For a suggested listing change, every field it would change sits beside the listing as it is now. One person can look after several audiences from the same app.

iPhone, iPad & Mac on theApp Store

Android is in testing, not on Google Play yet.

Arbiter needs an AudienceKit account with administrator access to an audience. Sign in with Apple or Facebook. Share audiencekit.com/apps/arbiter with your team.

The AI host, on your own Mac

An AI host with guard rails

On a Mac, Arbiter can do the work itself: it runs the audience’s decision worker and the AI host for venue chats on models stored on that Mac, so those messages aren’t sent to a third-party AI service.

It never knows who’s talking

The host reads the conversation with people replaced by letters. It never gets a member’s name, profile or tribes, and it only acts on messages from the live chat.

Code sets the limits, not the model

Whatever the model suggests, a fixed set of rules decides what may happen. Punitive calls always wait for a person, nothing targets staff, the venue or its hosts, and the host won’t chime in unasked more than once every two minutes in a room.

Automated bans don’t stick on their own

A ban proposed by automation goes to your staff straight away and lasts 24 hours at most unless one of your team confirms it. Every action is recorded with its reason, and any ban can be lifted.

Start in shadow

The host starts in shadow mode, where it carries nothing out and your team can judge each call it would have made. Move to approving every call, then to live, when you trust it. Each audience sets its own guidelines and thresholds in the admin console.

Honest about what it is

Asked, the host says it’s an AI and which model it runs. It answers from the chat and the venue’s own listings, and doesn’t invent events.

Your team sees its work

Every call the host makes is uploaded for review, with what it read and why it chose what it did. Anyone on your team can look from an iPhone or iPad without running a model.

For members

Every member can act

Safety starts with the people in the room, so the tools they need are one tap away in apps like Hot Mess.

Report anything

Report a message, a person, a venue, a ping or a photo. Each report keeps a copy of what was said, goes to the audience’s team in the admin console and to their phones in Focus, and reaches us too.

Block, both ways

A block hides two people from each other everywhere: in chat, in who’s here, in friends and in pings. Neither can reach the other, and the block is reported so the team can see patterns.

“I feel unsafe”

At a venue, a guest taps once and the staff on shift hear it straight away. The guest is told who is coming and can message them privately. Anything the app shows on the lock screen never says “safety” or “unsafe”.

For venues

Help on the floor

Safety alerts reach the people who can walk over, through Velvet at the door and behind the bar.

Staff on shift hear it

A venue’s Velvet devices with moderation access are on shift by default. Staff can choose to hear alerts only while they’re at the venue, so phones go on shift when they walk in and off when they leave.

One person picks it up

The first to tap “I’m on it” takes it, and the guest sees their name. Staff can message the guest, share support lines, and close the alert with a note.

Nobody answers? It goes up

After two minutes without a response the staff hear it again, the venue’s managers hear it too, and the guest is shown other ways to get help.

Teams can send practice alerts to rehearse the whole flow without anyone needing help. Alerts that look wrong go to the audience’s team for review, and only a person decides whether anyone is banned.

Arbiter on every device

Review anywhere. Host on a Mac.

One app, one sign-in, the same queue live on every device your team uses.

DeviceWhat it doesWhere to get it
iPhone & iPadThe three lists, deciding, the whole queue and every call the AI host made.App Store
MacEverything on iPad, plus the decision worker and the AI host on local models. It can keep working in the background from the menu bar.App Store (same purchase)
AndroidReviewing and deciding, like iPhone and iPad.In testing, not on Google Play yet
WebThe admin console: the reports queue, safety alert review, and the Moderation settings Arbiter follows.Your audience’s admin console

Child safety

Zero tolerance, handled for everyone

Every app on AudienceKit is for adults. Child safety is handled by the platform, so no community has to build it alone.

Known material only

Images are checked only against lists of known child sexual abuse material, never guessed at by a classifier. Members can report it from anywhere, with a reason of its own.

Withheld and preserved

A match or a report withholds the content from everyone at once and keeps a copy in restricted storage that only designated reviewers can open, with every look recorded.

A person files every report

A trained reviewer looks at every case and reports apparent abuse material to NCMEC’s CyberTipline, as US law requires. Read the Child Safety & CSAM Policy.

Your rules, your dials

The community sets the line

What’s banter in one scene is a problem in another. Automated moderation is something an audience turns on, not a default, and each audience sets it up in the admin console under Moderation:

  • Guidelines the host can post in a room when they’re needed.
  • The host’s mode: shadow, approve every call, or live.
  • Which rules run and how sure the checks must be before they act.
  • Which kinds of decisions the AI may take on, and which always wait for a person.
  • Any extra tools, like letting the host open a safety alert for a guest, are off until an administrator grants them.

See every safety tool on the integrations menu. To report a problem, use Report in the app, or write to rick@lovewins.media.

Your community, your call.

Keep your scene
safe and yours.

Tell us about the community you look after.
We’ll help you set up its rules and its host.

Start a conversation