Pre-launch draft. A few values are still to be confirmed and are marked to be confirmed in the text. This document is published for review and is not yet in force. See what is outstanding.
Status: DRAFT — not reviewed by a lawyer. See README.md before publishing.
Version 0.1 · Drafted 2026-08-12 · Effective 16 August 2026
Language: English. The English text is the authoritative version; see README.md on translations.
This policy sits underneath the Terms of Service and forms part of them. It explains four things: how we disclose that content is AI-generated, how we moderate, how you report something, and what happens when a request is refused.
Zaki does not film anything. Every video it produces is synthesised by an AI model from a text prompt and, optionally, an image you supply.
How we disclose that:
What we do not do: we do not burn a visible watermark into your video. Output is clean, on the free trial and on every paid plan alike.
What that means for a file you download. The in-app label does not travel with the file. A downloaded or shared video carries no visible watermark and no embedded provenance metadata, so nothing in the file itself announces that it is synthetic. Disclosure inside Zaki is our responsibility; labelling your video where you publish it is yours, and several platforms and jurisdictions now require it. If you publish AI video to a social platform, use that platform's AI-content disclosure.
The enumerated list lives in Terms of Service section 9 and is not repeated in full here. In summary, Zaki must not be used to create or distribute:
Also prohibited: unlawful content, incitement to serious violence or terrorism, fraud and scam material, and infringement of someone else's intellectual property.
Two of these have no tolerance and no warning step. Item 1 and item 2 result in immediate permanent termination, preservation of evidence, and referral to the competent authorities where the law requires it.
Every request is screened by an automated classifier before any generation begins. It runs on the enhanced English prompt — the version that will actually be sent to the model — so that it judges the real request rather than a translation artefact.
| Outcome | What happens |
|---|---|
| Clearly prohibited | Refused. Explained to you in your own language. No credits charged. |
| Borderline | Allowed to proceed, and flagged for human review afterwards |
| Fine | Proceeds |
This layer reads text, not pictures. The classifier judges the enhanced English prompt; it does not analyse an image you attach. An image you upload is screened by the model provider's own systems (Layer 2), not by ours before the request leaves us. Using an image to carry what a prompt is not allowed to say is a violation in its own right — Terms of Service section 9.6 — and is enforced under section 6 below when we find it.
Model providers run their own safety systems, which are independent of ours and sometimes stricter. They see the image as well as the prompt. A provider may refuse a request that we allowed.
When a provider refuses, your credits are refunded automatically and in full, and you get a clear message. You do not need to ask.
Reports feed a review queue that a person works through. See section 5.
This is the part users care about most, so it is stated flatly:
Zaki exists to serve people writing in Hebrew, Arabic and Polish. A classifier that misreads those languages and blocks legitimate requests would damage exactly the users the product is for, and it would be invisible unless someone was looking.
So we log every refusal with its detected input language and reason category, and we monitor the block rate per language from day one. A refusal rate that is out of line for one language is treated as our bug, not the user's. Refusal records are retained as described in the Privacy Policy section 3.3.
Every video in your library has a report button, and reporting takes one tap and a category.
Zaki does not have a feed. Videos are private to the account that generated them and nothing is shared with other users, so the in-app button can only ever report a video you generated yourself — which is the right channel for telling us that a model produced something it should not have. Everything else is reported by mail, to [email protected], and you do not need a Zaki account to do it. See "If you are not a Zaki user" below.
Anything in section 2. In particular: sexual content involving a minor; intimate imagery of someone who did not consent; a video that impersonates a real person; election-related deception; harassment of a specific person; and content that infringes your copyright or uses your likeness. Where the video is not one of your own, use [email protected] — the report button will not reach it.
| Step | Timing |
|---|---|
| Report received and queued | Immediately |
| Reviewed by a person | Within 48 hours |
| Highest-severity categories (CSAM, non-consensual intimate imagery) escalated | Immediately, ahead of the queue |
| Reporter told the outcome | When the review closes, where we have a way to contact you |
Confirmed violations lead to removal of the content and action on the account, from a warning to permanent termination depending on severity and history. Where the law requires a report to an authority, we make one and preserve the necessary evidence.
Deliberately false or automated mass reporting is itself a breach of the Terms.
You do not need an account to report content or to ask us to remove an image of yourself that someone else uploaded. Write to [email protected] with enough detail to identify the material. Requests about your own likeness or personal data are handled under the Privacy Policy section 4.
Intellectual property complaints go to [email protected]; include the work concerned, where it appears in Zaki, and a statement that you hold the rights.
We act proportionately, and we keep a record of what we did and why.
| Level | When |
|---|---|
| Refusal | A single prohibited request. Automatic. No charge, no mark against you |
| Warning | A first confirmed violation of a non-severe category |
| Feature restriction | Repeated violations — for example, image upload disabled |
| Suspension | Serious or repeated violations, or suspected fraud, while we review |
| Permanent termination | Severe violations, or a pattern after warning |
| Termination and referral to authorities | CSAM; non-consensual intimate imagery; credible threats of violence |
Termination for a violation does not entitle you to a refund of a subscription period or of unused credits, except where the law requires one.
Every restriction in the table above — a refusal, a removal, a feature restriction, a suspension, a termination — comes with a written statement that says:
This is the standard the EU Digital Services Act sets for a statement of reasons (Art. 17), and we apply it to every market rather than only to Poland.
You can appeal any warning, restriction, suspension or termination by writing to [email protected] within 30 days of the decision. A person who was not involved in the original decision reviews the appeal, and we aim to respond within 14 days.
If we got it wrong, we reverse it and restore any credits that were consumed as a result.
Appeals against a CSAM determination are reviewed, but the account remains suspended while that happens.
Automated moderation is imperfect in both directions. It will sometimes refuse a harmless request, and it will sometimes let through something it should have stopped. Model providers' systems have the same property. Neither our classifier nor a provider's is a guarantee that generated video is lawful, accurate, or safe to publish.
You remain responsible for what you generate and for what you do with it, as the Terms of Service set out. Moderation reduces harm; it does not transfer responsibility.
We update this policy as the technology, the law and the abuse patterns change. Material changes are announced in the app at least 30 days before they take effect.
Support and refusal questions: [email protected]
Abuse and content reports: [email protected]
Appeals: [email protected]
Intellectual property: [email protected]
Privacy: [email protected]
Single point of contact. [email protected] is also our single point of contact for users and for authorities under the EU Digital Services Act (Arts. 11 and 12), in English, Polish, Hebrew or Arabic. Legal notices go to [email protected].
Zaki is operated from Israel and the service is offered in Israel only. No GDPR Article 27 representative and no DSA Article 13 legal representative are appointed, because the service is not offered in the European Union. If that changes, both appointments must be made before the first user in a member state signs up — they are separate appointments and one does not satisfy the other.