AI and Avatar Policy

1. How our avatars work#

A Zentiq avatar combines several kinds of AI: speech recognition to understand what people say, language models to decide what to answer using the knowledge the customer provides, voice generation to speak, and video generation that animates a face in real time. We run our own video-generation models on servers we operate, and we use carefully selected AI service providers for speech, language and voice. Chloe, the host who helps customers build their avatars, is an AI avatar too.

AI can make mistakes. Avatars can misunderstand, answer inaccurately or say something unsuitable, which is why the safeguards below matter and why important decisions about people must stay with people.

2. Telling people it is AI#

• Every avatar is set up to tell people that they are talking to an AI. Wherever Zentiq shows an avatar, an AI label stays on it for the whole conversation, with text for screen readers, and a short notice says so when the conversation starts. The standard notice reads "You're speaking with an AI avatar." Customers may adapt the wording to their setting, but our Acceptable Use Policy forbids them from switching the disclosure off, hiding it or telling an avatar to claim to be human.
• Avatars are instructed never to claim or suggest that they are a person, including in role-plays, and to say that they are an AI whenever someone asks. Whether an avatar also says so in its first words is a setting for each organisation or subject; by default it does for learners under 18, job interviews and health, finance, insurance or complaints services.
• The pages we host for our customers' website visitors, job candidates and embedded avatars always state that the person is talking to an AI, and so does our support avatar. On our own website, Chloe is labelled as an AI.
• When a customer shows an avatar in its own app, our ready-made code shows the AI label on the avatar by default. It can be restyled or moved, and hidden only after an owner of the customer's organisation confirms that their app shows its own label for the whole conversation.
• Where an avatar resembles a real person, the customer must also make clear that the video and voice are AI-generated, as the law requires for such content.

3. Consent for faces and voices#

• A custom avatar or voice can be made from a real person only after the person adding the photo or recording confirms, in the product, either that it is their own face or voice and they agree, or that the person shown has given consent. For a duo, each person's consent is recorded separately.
• We keep a record of each confirmation: the exact wording and its version, who confirmed it and in what role, when and where, and a hashed IP address. Voice cloning is refused if no confirmation is recorded.
• Our terms require customers to hold written consent from every person depicted or voiced, and to show it to us if we ask.
• Consent can be withdrawn. When it is, the avatar or voice stops being used in new sessions straight away, and we delete the working copies used to generate it. Account owners can withdraw consent from the console. Anyone who believes their face or voice is being used without permission can write to legal@zentiqai.com with the subject line "Abuse report", and we will act on credible reports quickly, disabling the avatar while we look into it.
• Our library avatars are based on real people who agreed to their likeness being used in Zentiq.
• Faces generated by AI from a written description do not depict real people, and the person creating one declares this. We block generated avatars that look under 18.

4. Marking AI-generated video and audio#

The EU AI Act requires providers of AI systems that generate video and audio to mark their output in a machine-readable way, so that it can be detected as AI-generated. Sessions on Zentiq's own pages and in the website widget already tell people, on screen, that they are talking to an AI. Faces and characters designed with AI, duo pictures, AI-painted backgrounds and branded looks made in Zentiq now carry a machine-readable label in their metadata saying they were made or changed with AI (the industry-standard "digital source type"), and our API, webhooks and connections to AI apps mark the AI-generated content they return. We are still adding an invisible watermark in avatar speech and video, and signed content credentials on stored clips and images, with a free way to check them. We aim to switch these on across the Service before 2 December 2026, and we will update this page when they are live. Customers must not remove or interfere with any marking, watermark or label.

5. Human review in hiring#

• Our AI interviewer asks the questions the hiring organisation sets and rates the answers against its criteria, to help its reviewers. It never rejects anyone.
• Only a person on the hiring team can record a decision. A candidate's interview must be opened or reviewed before a decline can be recorded, and candidates cannot be declined in bulk.
• Before the interview, candidates are told that they are talking to an AI, how their answers will be used, and, where recording is switched on, that the interview is recorded; they must agree before they start. For roles in places with specific rules, such as New York City and Illinois, the product includes notices written for those laws; the hiring organisation remains responsible for giving every notice the law requires, at the time it requires.
• Hiring organisations remain responsible for their decisions, for any bias audit the law requires, for offering alternatives and adjustments, and for complying with employment and anti-discrimination law.
• During an interview, software on the candidate's device counts the faces in view to help keep the interview fair. If the hiring organisation records the interview, with the candidate's agreement, the recording is kept as our Privacy Policy describes. We do not analyse candidates' emotions.
• Interview data is deleted automatically at the end of the retention period the organisation sets for the role, subject to the minimum periods that employment law requires in some places.

6. Protecting children#

• Zentiq accounts are for adults. Children take part only as learners in schools, tutoring services and family accounts.
• A child's first session is blocked until consent is on file from a parent or guardian or, where the law allows, from the school for its educational purposes.
• For children, conversations are kept as transcripts only by default, camera tips are never offered, and our staff must confirm again before opening a transcript from a school or family account.
• Notices shown to children use simpler words, and learners' conversations are checked for signs that someone may be at risk, so that the organisation's safeguarding contacts can follow up.

7. Cameras and emotions#

• Camera features are optional and are analysed on the person's own device; the video is not sent to us, except when an organisation records a job interview with the candidate's agreement.
• Expression reading is available only in the console's test area, where people try out their own avatars. It is off by default, is switched on only by the person in front of the camera, and lets the avatar respond more naturally. It is not used in learner sessions, website conversations, job interviews or assessments.
• Camera tips in workplace training measure delivery, such as eye contact with the camera, staying in frame and movement, to help learners practise. The learner chooses to switch them on, only the resulting numbers are kept, and they are never offered to children or in hiring or assessments.
• The EU AI Act prohibits AI that infers the emotions of people at work or in education, except for medical or safety reasons. Customers must not use any Zentiq feature to infer the emotions of their employees, job candidates or students where that law, or any similar law, applies.

8. Your data and our models#

We do not use customers' content or conversations to train AI models that are made available to other customers, and we never sell them. Face and voice data is used only for the avatar or voice it was provided for. Our Privacy Policy explains how long we keep data and how it is deleted.

9. Laws we follow#

• EU AI Act. We act as the provider of the AI system behind Zentiq avatars; customers who use them act as deployers. The ban on certain practices, including emotion recognition at work and in education, has applied since 2 February 2025. The duty to tell people they are interacting with AI has applied since 2 August 2026, and machine-readable marking of AI-generated content is required for systems already on the market by 2 December 2026. Uses such as recruitment and education can be "high-risk" under the Act; those obligations apply from 2 December 2027, and we are preparing the documentation and oversight features that customers will need.
• United States. We design the hiring features to support the New York City law on automated employment decision tools, the Illinois Artificial Intelligence Video Interview Act and Illinois Human Rights Act rules on AI in employment, California's rules on automated decision systems in employment, and Colorado's law on automated decision-making technology when it takes effect on 1 January 2027.
• Privacy and biometric laws in Europe, the UK and the US, as described in our Privacy Policy.

10. Reporting misuse#

If you see a Zentiq avatar being used to deceive people, to impersonate someone, without the consent of the person it depicts, or in any other harmful way, please tell us at legal@zentiqai.com with the subject line "Abuse report". Our Acceptable Use Policy explains what to include and what happens next. Privacy questions go to privacy@zentiqai.com.

11. Keeping this policy current#

AI and the law around it are changing quickly. We review this policy at least once a year, and whenever our product or the law changes in a way that affects it, and publish each new version here.