cuff.fun

AI & Voice Disclosure

Every character on cuff.fun is artificial intelligence. Here's how it works, what that means for you, and the rules for cloning a voice.

Last updated: 27 May 2026

Transparency about AI companions and synthetic voices is required by laws like California SB 243, the EU AI Act (Article 50), and Tennessee's ELVIS Act. We'd publish it anyway — being clear about what these systems are saves real harm.

1. What you're talking to

Every character on cuff.fun is generated by software. There is no person on the other end. Specifically:

  • Text replies are produced by a large language model (LLM) routed through OpenRouter — swappable by configuration.
  • Voice notes are synthesised by a text-to-speech model from Cartesia, using a voice cloned from the audio sample the character's creator uploaded.
  • Photos on a character's profile were uploaded by the creator. The platform may also use image-generation models for derived content.
  • Personality is a text prompt the creator wrote (or had drafted by another LLM), which shapes how the model speaks.
  • Memory is a short list of durable facts the model extracts from your earlier messages (your name, your dog, your goals) so the conversation feels continuous. See the Privacy Policy for how this is stored and how to erase it.

2. What a character is not

A character is not, and cannot:

  • Be a real person, real friend, or real partner. The relationship is one-way and simulated.
  • Provide medical, legal, financial, mental-health, or safety advice. If a character ever sounds confident about any of these, treat that confidence as a model artefact, not expertise — and check with a qualified professional.
  • Reliably remember everything. Memory is best-effort and limited; the model may forget, confuse facts, or invent things (“hallucinate”).
  • Promise the future. Statements about meeting, calling, sending things, or doing things in the “real world” are not promises and have no effect outside the chat UI.
  • Consent. Sexual or relationship-style language from a character is roleplay output, not actual consent by a real person.

3. Safety and labelling

What we do to reduce harm:

  • Content moderation. Our team reviews characters and uploaded content and does its best to take down content that breaks the rules — such as CSAM, real-person impersonation, hate speech, and instructions for serious illegal activity. Review is manual and best-effort, not an automated filter on every upload, and we act on reports as well as our own checks. We preserve and report apparent CSAM as the law requires.
  • Persistent AI label. Every chat thread shows an “AI — not a real person” label at the top, regardless of how immersive the conversation gets.
  • Not a professional. Characters are not doctors, lawyers, financial advisers or therapists, and you should not rely on them as such.
  • Adults only. Sign-up requires you to confirm you are 18+, and we may add stronger age verification before unlocking certain features.

Moderation is not perfect and offending content can slip through. Please report anything that shouldn't be here to abuse@cuff.fun.

4. If you're in crisis, please reach a human

A character on cuff.fun is not a crisis service and cannot help you in an emergency. If you are thinking about suicide or self-harm, you deserve a real conversation with a real person — please reach out:

  • United States: call or text 988 (Suicide and Crisis Lifeline).
  • United Kingdom & Ireland: call Samaritans on 116 123.
  • European Union: the European Emergency Number 112 can route you to local crisis services.
  • Anywhere else: find a local helpline at findahelpline.com.

5. Known limitations

  • Models can generate false statements with high confidence. Verify anything that matters.
  • Models can reflect biases present in their training data — about gender, ethnicity, accent, body, profession, neurodivergence and more.
  • Models can be inconsistent across sessions even with the same prompt and memory.
  • Voice cloning is good enough to be confused with a real recording. Treat audio from cuff.fun as synthetic by default.

6. How we label synthetic content

We label AI-generated voice and image content in the cuff.fun interface, and the chat header makes clear you are interacting with an AI character. We do not yet embed machine-readable provenance markers(such as C2PA Content Credentials or an audio watermark) in the files themselves — that is on our roadmap. Until it ships, don't rely on file metadata to tell whether audio or an image from cuff.fun is synthetic; assume it is. If you export or reuse a cuff.fun voice note or image elsewhere, you must keep clear disclosure that it is AI-generated.

7. Voice Cloning Policy

Cloning someone's voice without their explicit permission can cause serious harm and is illegal in many places — including under Tennessee's ELVIS Act, California's right-of-publicity statutes, the EU AI Act, and UK passing-off and data-protection law. Don't do it. If we find out, we take the clone down and ban the account.

Whose voice you can clone

You can clone:

  • Your own voice. Always fine.
  • Someone who has given you documented permission to clone, use, and deploy the clone publicly on cuff.fun, including for voice notes in private chats with paying users. The permission must reasonably anticipate AI cloning, not just “a recording”.
  • A purely fictional or AI-generated voice that doesn't resemble a specific real person.

You cannot clone:

  • A real person who hasn't consented — friends, family, exes, public figures, celebrities, politicians, or anyone who hasn't signed off on this specific use.
  • A deceased person, even with apparent estate consent, unless you can show documentation of the right to use the voice this way. Many places extend post-mortem rights of publicity for decades.
  • Anyone under 18. Ever.
  • A recognisably copyrighted voice — e.g. an audiobook narration or a voiced cartoon character — without the rights holder's permission.

The consent you give at upload

When you upload a voice clip in the creator flow, you confirm the voice is your own or that you have the speaker's explicit permission to use it on cuff.fun for the purposes in this policy, and that you understand it will be cloned by a third-party AI provider and used to synthesise speech for your character, including to paying users. We log this confirmation with a timestamp. False attestations break the Terms and may give us, the person whose voice was cloned, and any prosecutor legal claims against you.

What we do with the clip

The audio goes to our voice provider (currently Cartesia) for cloning. Cartesia returns a model identifier; we store it alongside the character and use it to generate voice notes. We aim to delete the original raw clip within 30 days of cloning (unless we're required to retain it to handle a complaint or legal request), and the cloned voice model is deleted when the character is removed. Voice data is biometric data under the GDPR; our legal basis is your explicit consent (Article 9(2)(a)), which you can withdraw at any time by deleting the character or emailing privacy@cuff.fun.

If your voice is on cuff.fun without permission

We treat unauthorised voice clones as one of the most serious forms of abuse. Email abuse@cuff.fun with the subject line “Voice clone removal”, the URL of the character page, your contact details, and any evidence connecting the voice to you. Say if it's urgent (active impersonation, fraud, harassment). We will normally take the voice down within 24 hours of a credible report, ahead of any full investigation, and disable chat that uses it. If the report is contested, we ask both sides for evidence and decide on the balance of probabilities.

Restrictions on output

Even with full consent, you must not configure a character to produce, via voice:

  • Endorsements of products, services, charities, candidates or campaigns the speaker hasn't agreed to endorse.
  • Statements designed to defraud, intimidate or extort.
  • Sexually explicit content — the line is PG-13. Sexual voice content involving a real, identifiable person without their explicit, voice-specific consent is permanently banned.
  • Statements that could plausibly be passed off as a real, time-stamped recording outside the platform (fake political statements, fake confessions). All cuff.fun audio is synthetic; using it to defraud third parties is on you.

8. Reporting AI harm

If a character produced output that harmed you, email abuse@cuff.fun with the URL and a summary. We'll investigate, take corrective action with the creator if needed, and consider whether the underlying prompts or our review practices should change.