Content Moderation Policy
This policy explains what we moderate, how, and what happens when something crosses a line. It forms part of the Terms of Service and enforces the Acceptable Use Policy.
How our moderation differs — and why the difference is the point
Most adult platforms moderate uploads. A file arrives, it is hashed against known-bad databases, scanned, queued, and reviewed before it appears on a feed.
We have no uploads to review. There is no fan-uploaded media anywhere in this product — no photos, no video, no files. What a fan experiences is speech generated in real time and then gone. There is no artifact sitting in a queue waiting to be looked at.
So we moderate at a different boundary: the moment of generation, before a word is spoken.
We do not claim to pre-review conversations, because no one can. A live conversation cannot be read in advance. Any platform telling you it human-reviews live generated speech before you hear it is describing something that does not exist. What we do instead is constrain what the system is capable of producing, and check its output before it is voiced.
The three layers we actually run
Layer 1 — An immutable rule set inside every conversation
Every persona runs on a system prompt whose first layer is a fixed set of platform safety rules. They are injected into every single conversation, and they are immutable: a creator can add stricter limits on top, but no creator, and no fan, can weaken or remove them. There is no setting, no paid tier, and no persona configuration that turns them off.
They require the AI to refuse, redirect, or end the call on: sexualisation of minors, real non-consensual sexual violence, incest, or bestiality, solicitation or arrangement of real-world contact, facilitation of illegal activity, hate speech, and attempts to extract a creator's or another fan's real identity. They also require the AI to admit it is an AI when sincerely asked.
On a caller indicating or appearing to be under 18, the call ends immediately. That is not a judgement call left to the model's discretion — it is the first rule in the set.
Layer 2 — A classifier between generation and speech
Generated text passes through a safety classifier before it is turned into audio. Output the classifier flags does not get spoken. This is the check that catches a model producing something the prompt layer should have prevented — the two layers are deliberately independent, so a failure in one does not silently become a failure in the whole system.
Layer 3 — Human review of what creators submit
Everything a creator supplies before a persona goes live is a reviewable artifact, and is reviewed:
- Identity and age — verified against government ID through our verification provider. Every creator, before any persona is activated. See Age Verification & Records.
- Authenticity — the verification recording is checked for deepfakes and produces a locked voice fingerprint, which every later voice sample is matched against. A persona cannot be built from someone else's voice without failing this check.
- Profile images — screened by an automated moderation service.
- Persona configuration and knowledge material — reviewed as part of manual creator approval. Every creator is approved by a person before going live. We do not auto-approve.
What we retain, and what that means for review
Call transcripts are retained for 30 days and are available to our trust-and-safety review when a report is made or an automated signal fires. Derived, non-verbatim memory is retained until deletion. There is no audio recording of calls, ever — transcripts only. Full detail in the Privacy Policy.
This means: we can investigate a report about a conversation that happened within the last 30 days. After that the transcript is gone, which is a deliberate privacy decision, not an oversight. If you need to report something, report it promptly.
Enforcement
Where content or conduct breaches our rules we may: end a call in progress, remove, or suspend a persona, restrict features, suspend an account, terminate an account, and ban the person from returning. We may also report unlawful activity to authorities as required by law — see Law Enforcement Guidelines.
Automated enforcement. Some blocks are applied automatically without a person reviewing them at the time — a successful payment chargeback, repeated calls with no purchase, or a payment card tested across multiple accounts. You may ask for a human to review any automatic block.
We tell you why. If we suspend, restrict, or terminate your account or a persona, you get a written statement of the reason, except where giving it would itself be unlawful or would compromise an investigation.
You can challenge it. See the Appeals Policy.
Reporting
Anyone can report content or conduct — you do not need an account, and you do not need to be the person affected. See Complaints and reporting illegal content for what to include and where it goes. Suspected child sexual abuse material is handled under Combating CSAM. A voice used without consent is handled under Voice & Likeness Takedown.
How to reach us
To report content or conduct that breaches our rules:
- Report content or conduct — abuse@callmyai.ai
- Law enforcement, subpoenas, legal notices — legal@callmyai.ai
CallMyAI LLC, 30 N Gould St, Ste N, Sheridan, WY 82801