Verify AI-generated content

1. What we mark

• Website text. Copy for customer websites is generated with Anthropic Claude models, which embed an invisible statistical watermark in the wording itself. It survives copying, HTML and code export, and contains no extra characters.
• Pages. Every published or exported page declares <meta name="ai-disclosure" content="ai-autonomous" ai-provider="TrueOwned (Lynx Group GmbH)" ai-prompt-url="https://trueowned.com/verify" /> (W3C AI Content Disclosure). Exports and repositories also contain an AI-DISCLOSURE.md file.
• Images. AI images are generated with FLUX.1 on fal. The original file carries fal's signed C2PA Content Credentials and fal's invisible watermark. Every resized or converted copy we publish carries IPTC metadata DigitalSourceType = http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia and a C2PA manifest signed and time-stamped by Lynx Group GmbH that links back to the original. We never publish AI copies narrower than 800 px, so the watermark stays detectable.
• Visible labels. Site owners can show an “AI-generated image” caption on any generated image.

2. Check an image

1. Download the image (right-click, save) and open contentcredentials.org/verify. It reads the C2PA manifest in your browser: signer Lynx Group GmbH, the action “resized” or “transcoded” with source type “trained algorithmic media”, and the fal-signed original as ingredient.
2. To check the invisible watermark, upload the image to fal.ai/verify. It also detects copies whose metadata was stripped, for example screenshots or re-encodes.
3. Viewers report our signature as “untrusted” because our certificate is not yet on the C2PA trust list. Compare it with our root certificate below; a manifest that chains to it was signed by Lynx Group GmbH.
4. Command line: c2patool image.webp prints the manifest; exiftool shows the IPTC DigitalSourceType.

3. Check website text

Anthropic's text-watermark detector is not public; Anthropic gives access to organisations that must verify watermarking. Our application is pending. Until it is granted, send the page URL (or the text) to hello@trueowned.com with the subject “AI content verification”. We confirm from our records whether the text was generated with TrueOwned and with which model, and we run the detector once we have access. Short passages (headlines, buttons, under about 200 words) cannot be detected reliably.

4. Our signing certificate

The root certificate is published here once signing is configured for this environment.

5. Do not remove the marks

Our Terms prohibit removing or altering these marks. If you find TrueOwned content with the marks removed, or a mark you believe is wrong, tell us at hello@trueowned.com.