Legal

AI Terms

Last updated 24 September 2026 · Version 2026-09-24

These terms apply when you use OpusFunnel's AI features. They add to our Terms and Acceptable Use Policy. The short version: AI output can be wrong, people must know when they are dealing with AI, and you are responsible for what you publish and send.

Which features use AI

Who receives what

• Anthropic (Claude models) runs most text features. It receives the prompt a feature builds, which can include contact details, message history, review and comment text, form answers and your business information.
• OpenAI creates and edits images (it receives the pictures and logos you upload), turns text into search vectors for the knowledge base and reply suggestions (it receives message text, callers' questions and your documents), and runs some text generation.
• Telnyx runs the AI phone agent: speech recognition, the language model (which Telnyx selects) and the voice. It receives the call audio and the details the agent needs.

Before they build a prompt, the AI agents drop values that look like card or social security numbers, and custom fields with names like "password". The providers are listed on our sub-processor page.

Training

We do not train AI models on your data or your contacts' data. Under their business terms, Anthropic may not train models on data sent through its API; OpenAI does not use API data to train or improve its models unless the customer opts in; and Telnyx does not use it to train models offered to others without the customer's written opt-in, except in de-identified form.

Providers keep some data for a limited time. Anthropic deletes API inputs and outputs within 30 days, but keeps content flagged for breaking its usage policy for up to 2 years. OpenAI keeps abuse-monitoring logs, which can contain prompts and responses, for up to 30 days, unless the law requires longer or it is needed to prevent harm. Telnyx keeps call transcripts and recordings for as long as our settings with it say; when you delete a recording in the app or close your account, we ask Telnyx to delete it.

Telling people it is AI

• When the inbox assistant replies to a contact automatically, or a workflow agent sends a contact a text or email, the first such message to that contact on that channel, and the first again after 30 days, says it comes from your business's automated assistant and how to reach a person (on a text, reply HUMAN; on an email, reply to it). The AI phone agent says it is an AI assistant at the start of each call. These disclosures cannot be switched off.
• Messages sent by workflow agents go out in your business's name. Do not present them as written personally by a member of your staff.
• Do not tell an agent to claim to be a person, and do not deny it is AI if someone asks.
• Replies to reviews and social comments are posted in your business's name. If automatic replies are used to sell to people, the law in some places (such as California) requires you to say a bot is involved.

AI images

• Images made with the image ad and photo tools are AI-generated. OpusFunnel writes a machine-readable "made by AI" mark into each PNG, JPEG or WebP file they produce (the IPTC digital source type "trainedAlgorithmicMedia", or its composite form when text or a logo is laid over the picture), unless the file already carries a signed provenance record from the model that made it. The mark says how the picture was made and does not name OpusFunnel. When OpusFunnel posts one of these images to Instagram, X or Pinterest, it also turns on that network's AI label. A visible label is still your responsibility wherever the law or a network requires one.
• Realistic images of people, places or events that did not happen must be labelled as AI-generated when you publish them, and some ad networks require an AI flag on upload.
• Only upload pictures of people who have agreed, and never make an image of a real, identifiable person without their consent.
• Image ads can quote your published reviews with the reviewer's name. Only use reviews you have the right to use in advertising.

Accuracy and review

AI output can be wrong, out of date or made up, even when it sounds confident. It can misread a question or a caller. Check copy, prices, claims and legal statements before you publish them. Keep a person responsible for automatic replies and read what they send. Use the knowledge base to give agents correct facts, and switch off automatic modes you cannot supervise.

What AI must not decide

Do not use AI lead scores, agent replies or any other AI output to decide whether someone gets credit, a job, housing, insurance, education, healthcare or another essential service. Lead scores are a sales aid. If you use them in a way that has legal or similarly significant effects on people, you must meet the rules on automated decision-making (such as Article 22 of the GDPR), including human review.

Your responsibility

• You are responsible for the prompts and material you give AI features, and for everything you publish or send, whether a person or AI wrote it.
• You must have the right to use what you upload, and a lawful basis for the personal data you let AI features process.
• Tell people in your own privacy notice that you use AI on their messages and calls.
• Do not use AI features for anything the Acceptable Use Policy or the providers' usage policies prohibit.
• If your staff use AI features, make sure they understand what the features do and their limits.

Charges and controls

AI use is charged to your balance at the rates shown in the app. When your balance runs out, AI features stop. Agencies can switch AI off for each client account.

Laws on AI

AI rules are changing quickly. They include the EU AI Act's transparency duties, the FCC's ruling that AI voices in calls are artificial voices under the TCPA, and state laws on bots and automated decisions. You are responsible for following the rules that apply to how you use AI. We will update these terms and the product as the rules change.

Version history: 2026-09-24, first version.

Contact