Acceptable Use Policy

Last updated

Voice agents and synthetic speech are powerful. This policy is part of our Terms of Service and applies to every agent, session, voice, request and account. It includes the use restrictions that come with the speech and language models we build on. If you're unsure whether a use is allowed, ask us first at .

1. Never build agents or generate speech that

• Impersonate a real person without their permission, suggest a real person said something they did not say, or pretend to be a specific business, government agency or official you don't represent.
• Claim to be human when asked sincerely, or hide that they are automated where the law requires disclosure.
• Make unsolicited or illegal calls: robocalls, telemarketing without the required consent, calls to numbers on do-not-call lists, or any calling that breaks telemarketing or robocall laws (such as the TCPA in the United States).
• Deceive or defraud: scams, phishing, collecting passwords or payment details under false pretenses, fake emergencies, or "voice of a relative" schemes.
• Defeat security: bypass voice biometrics, voice-based identity verification or authentication systems.
• Harass, threaten, bully or intimidate anyone, or promote violence or hatred against people based on protected characteristics.
• Produce sexual content involving minors, engage minors in sexual or romantic role-play, or produce sexual content using a real person's voice without their consent. We report child sexual abuse material to the authorities.
• Interfere with elections or civic processes, including voter suppression, impersonating candidates or officials, or deceptive political content.
• Record or transcribe people covertly, without the notice and consent the law requires where they are.
• Break the law, including laws on privacy, publicity rights, consumer protection, telemarketing, discrimination and intellectual property.

2. Voice cloning rules

• Only clone your own voice, or a voice whose speaker gave you clear, informed, written permission covering how you'll use it, including in a voice agent. Keep that permission and produce it if we ask.
• Never clone a public figure, celebrity or performer without documented authorization, or a minor without verifiable parental or guardian consent.
• Never clone a voice from recordings of a person who doesn't know about it, such as phone calls, voicemails, videos or podcasts.
• If the speaker withdraws permission, delete the voice and stop using it in your agents.

3. Uses that need care

• No emergency services. Don't use an agent to answer or route emergency calls, or as the only way someone can reach help.
• High-stakes decisions need a human. Don't let an agent make final decisions about medical care, legal matters, credit, insurance, employment, housing or access to essential services without meaningful human review.
• Not professional advice. Agents that discuss health, legal, financial or mental-health topics must make clear they are not a licensed professional and point people to qualified help, including crisis resources where appropriate.
• Companion and role-play agents must not target minors, and must not encourage self-harm, isolation or dependency.

4. Disclosure and consent

• Tell End Users they are talking to an AI agent at the start of the conversation wherever the law requires it, and whenever they could reasonably believe they are talking to a person. Where a listener could believe they're hearing a real person's voice, or where the law requires it (for example, automated calls or political advertising in many places), clearly disclose that the voice is AI-generated.
• Tell End Users the conversation is transcribed (and stored, if your retention setting keeps transcripts), and get consent wherever the law requires it. Some places require every participant's consent.

5. Platform integrity

• Keep API keys on your server. Give browsers and apps only the short-lived session tokens the API issues.
• Don't put direct identifiers (emails, phone numbers) in customer_id, and don't put secrets in prompts or tool definitions that the agent could repeat.
• Don't create multiple accounts to collect free credit or get around limits or enforcement, and don't attempt to extract model weights, other customers' voices or data, or to probe, overload or disrupt the Service.

6. If you build on us

If you let your own customers create or configure agents or voices through your product, you're responsible for their use too. Pass these rules on to them, and have a way to handle reports.

7. Enforcement

We may stop sessions, disable agents, remove content or cloned voices, revoke keys, suspend or close accounts, and report illegal activity to law enforcement. For serious violations such as fraud, illegal calling, non-consensual impersonation or child safety, we act without warning and forfeit remaining credit. We may share relevant records, including consent records, with law enforcement or with a person whose voice was misused, where the law allows.

8. Report misuse

Got a call or message from a Ciabatta agent that broke these rules, or heard your voice where it shouldn't be? Report it here or email with the date, time and, if you have it, the phone number or website involved. We review every report.