Transparency about the use of artificial intelligence

CLAQON uses artificial intelligence in several parts of video production, but it does not let a model decide on its own what gets published. This page explains where AI is involved, which controls go with it, how it helps you disclose synthetic content and what is up to you.

Where CLAQON uses AI

You choose the text model from among the preconfigured providers, a local model or a compatible endpoint. The other paid services are only used if you enable them and with your own keys. More details in AI models and voice and visuals.

Human and deterministic controls

• Evidence first. Before writing, CLAQON researches with citable sources or with yours. The script can only rely on verbatim excerpts from the source.
• A deterministic check decides. HARD QC is not an AI model: it rejects figures that are not in the evidence, unsupported absolutes and unsourced claims. The script moves on at 100/100; if it fails, a targeted repair runs and, if it still fails, it goes to human review before anything is spent on voice or visuals. See source verification.
• The scoring is not done by AI. Each topic’s opportunity is scored with fixed rules, and learning only adjusts it within a margin.
• Continuous improvement has limits. It only touches safe settings, such as the caption style or the colour treatment, and reverts them if results get worse. It never changes the voice, schedules, budget, platforms or length on its own: those only reach you as a proposal.
• You decide when something is published. You can check the configuration without spending, create a draft without publishing or stop an entire account, with confirmation.
• No extra paid services without your permission. Paid voice, image and video are off by default, and you can set spend caps per video, per day and per month.

These controls reduce errors, but they do not guarantee there will be none. Review what you publish.

How it helps you disclose synthetic content

• YouTube. When any scene in a piece is a generated image or video, CLAQON flags the upload as synthetic content (the containsSyntheticMedia field of the YouTube API). Synthetic voice on its own does not trigger the flag, because YouTube does not require disclosure of synthetic narration that does not purport to be realistic. CLAQON can upload the video as private and make it public once the disclosure has been accepted.
• Provenance record. Each piece has a downloadable record with the music and effects and their licences, the clips and their providers, the credits and whether there was any generated image or video, with the digital source type according to the IPTC vocabulary (DigitalSourceType), in machine-readable form. Anything not on record is left blank; it is never filled in.
• Commentary Room (beta). It discloses all narrated commentary as synthetic.
• AI characters (beta). It discloses the use of AI on YouTube, adds an AI caption and the #AI tag on Instagram, and supports an on-screen badge.
• Real people. The real-people guard warns you if a generated scene mentions someone by name.

More details in compliance record and in the guide to labelling AI-made content.

What CLAQON does not do yet

• It does not yet embed C2PA or XMP metadata in the final video. The provenance record is kept separate. It will do so before it is placed on the market, as explained in the next section.
• It does not replace each platform’s label. The AI caption does not replace the disclosure tools of Instagram or Facebook, and on X CLAQON does not apply any AI label.
• It does not prevent platform labels. YouTube and Meta may label a video on their own initiative, for example when they detect metadata or on sensitive topics. Nobody can promise a video without labels.
• It is not intended for creating deepfakes. CLAQON is not designed to make real people appear to say or do something, and using it for that purpose is prohibited by the acceptable use policy.

Article 50 of the AI Act

Regulation (EU) 2024/1689 on artificial intelligence (the AI Act) sets out transparency obligations in its Article 50, applicable from 2 August 2026:

• Providers (Article 50(2)). Anyone who offers a system that generates synthetic audio, image, video or text must mark its outputs in a machine-readable format that allows them to be detected as artificial. Regulation (EU) 2026/1744 gives systems placed on the market before 2 August 2026 until 2 December 2026 to comply.
• Deployers (Article 50(4)). Anyone who uses the system must disclose, visibly or audibly, deepfakes and generated text published in order to inform the public on matters of public interest. The latter is not necessary if the text undergoes human review or editorial control and a person holds editorial responsibility.

How this is split in CLAQON.

• You deploy. As the operator of your channels, you decide what is published: the visible disclosure and labelling of what you publish are up to you, under the terms of Article 50(4).
• We will be the provider. We understand that, by placing CLAQON on the market under its brand, its owner will be a provider for the purposes of Article 50(2), even though third-party models generate the voice, image or video. That is why CLAQON will not be placed on the market without marking its synthetic outputs in a machine-readable format that allows them to be detected, as that article requires.

How we use AI on this website

• Texts. We use AI tools to help us draft and translate some of the texts on this website. We undertake that a person will review them before they are published and take responsibility for them.
• Panel screenshots and tour. They show the real panel with fictitious data from a sample environment: it is not customer data.
• No automated assistants. The website has no chats or AI assistants that converse with you.

If you think anything on this page is inaccurate, write to us at soporte@claqon.com.