Moderation Policy
Effective 3 November 2026 · Version 2026-11-03.1
Contents
This policy says what we check on Flirty, how, and where our checks stop. It supports the Acceptable Use and Content Policy. We would rather say plainly what we do than promise more.
Checks reduce risk, they do not remove it. We use automated checks and human review, and we say below where each one stops. Anyone can report what we miss.
1. What we check
| Where | What happens |
|---|---|
| When a persona is made | A persona card is rejected if its text, tags or name state an age under 18 or use words that suggest a minor. Only adult personas can be created and published |
| What the text model is told | Every conversation carries firm limits: every person is an adult, nothing sexual or suggestive involving a minor, no real identifiable person, nothing degrading a group, nothing illegal. Explicit talk between adults is allowed inside these limits. These limits are instructions to the model. They are not a filter that can be proven to catch every message |
| Before a picture or clip is generated | A request for media that names a minor (for example with the word "teen") is refused before anything is produced, and the tokens are returned. This is a check of words, so it catches what it names and no more |
| After a photo is generated | An automated check works out how much a photo shows (the level it falls into) and looks for obvious faults, such as extra limbs. The level is used to make sure the photo matches what was offered. This check does not tell the age of a person in the picture or whether a face matches a real person. It is recorded for video clips, not yet enforced |
| After publication | Anyone can report. Staff review reports under the Complaints and Content Removal Policy |
2. Humans in the loop
Staff review reports, appeals and cases of suspected abuse or fraud. A person decides to close an account. Staff may read part of a conversation to check quality and safety of what the service produced, to look into a report or abuse, or to fix a fault. Each time staff open a page of a conversation, we log who did it and when. We do not read conversations to judge you, and we have no view on what adults talk about within the rules.
3. What we do not do
- We do not claim to scan every message, picture or clip. There is no general duty on us to search the service for illegal material, and nothing here creates one.
- We do not detect minors or real persons in generated pictures with technology today. We prevent them by how personas and prompts are made and by reports.
- We do not remove content only because someone dislikes it.
4. When we get it wrong
- Something was removed or refused that should not have been: appeal it (Complaints and Content Removal Policy, section 6).
- Something is on the service that should not be: report it. We act as the policy says.
When we are unsure about content that may involve a minor, a real person or a non-consensual act, we remove it.
5. Improving
We review how the checks perform and tighten them as we learn. We update this policy when a check changes.