This document describes how UAB AIVON ("Aivon") maintains human oversight over its AI-generated content pipeline. It is published in accordance with emerging EU AI Act obligations applicable to providers of AI-generated content tools, and reflects our internal commitment to responsible operation.
Fully automated AI systems must have meaningful human checkpoints. This document explains what those checkpoints are, who is responsible, and what happens when a problem is identified.
This process covers all AI-generated content produced through the Aivon platform, including:
It does not cover content uploaded by users as reference inputs, which is governed by our Privacy Policy and Terms of Service.
Before any image is generated, the following automated controls are applied:
User-submitted prompts are evaluated for content that violates our Ethical Framework. Prompts containing keywords or patterns associated with prohibited content categories - including CSAM, non-consensual imagery, and incitement to violence - are blocked before reaching the model.
The platform exposes a limited set of style options and generation parameters. Users cannot directly access raw model parameters in ways that could be used to circumvent safety guidelines.
All generated images pass through an output review step that checks for content inconsistent with the submitted prompt or that triggers safety classifiers. Images flagged by this step are not delivered to the user and are queued for human review.
Flagged images are reviewed by a designated member of the Aivon team within 72 hours of flagging. The reviewer determines whether the image:
Any user or third party who believes a piece of content generated by Aivon violates our policies or applicable law can submit a report via our Contact page with the subject line "Content Report".
Reports are acknowledged within 48 hours on business days. Investigation and resolution target time is 7 calendar days for standard reports. Reports involving CSAM or imminent harm are treated as critical and escalated immediately (see Section 6).
Reporters will receive a confirmation that their report has been received and a final outcome notification once the review is complete.
The following categories are escalated immediately outside of normal review cycles:
| Role | Responsibility | Response Time |
|---|---|---|
| Platform Administrator | Day-to-day content review, user warnings, account suspensions | 72 hours |
| Management | Escalated cases, legal notices, CSAM reporting | 24-48 hours |
| Legal Counsel | Court orders, regulatory correspondence, complex IP disputes | 48 hours |
All content review decisions are logged internally with the following information:
These records are retained for a minimum of 3 years and are available to regulatory authorities upon lawful request.
Aivon reviews the effectiveness of its oversight process quarterly. Metrics tracked include: volume of flagged content, false positive rate, average resolution time, and user report outcomes. Findings inform updates to prompt filters, model selection, and this policy.
To submit a content report or ask questions about our oversight process, contact us at [email protected] or via our Contact page.
UAB AIVON
Pelesos g. 17-14, Vilnius LT-02114, Lithuania
Company code: 307619983