Brief IA

Anthropic Launches a C2PA Verifier, Without Text Support

🤖 Models & LLM·Tom Levy·

Anthropic Launches a C2PA Verifier, Without Text Support

Anthropic Launches a C2PA Verifier, Without Text Support
Key Takeaways
1Anthropic offers a free tool to check if image, audio, or video files bear traces of processing by Claude via the C2PA standard
2Text files are not publicly supported; text detection goes through a separate API in private preview
3Watermarks embedded in Claude texts are claimed to be invisible and have no impact on quality or cost
💡Why it mattersThe tool marks a step towards the traceability of AI-generated content, but its scope remains limited to non-text media for the public.
Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

📄
Full Analysis

Anthropic offers a free online tool to detect whether images, sounds, or videos show signs of having been processed by Claude. The company relies on the C2PA standard and promises an extension if it covers other formats. Textual content is currently excluded from the public service and is processed through a private detection API.

Limited Formats and Possible False Negatives

Anthropic's verifier only supports a limited number of formats, and no text files are accepted at this time. The compatible file types are multimedia formats such as JPEG, GIF, MP3, and MOV. When no signs of processing by Claude are detected, the service displays a message stating that no clues have been found, but this does not rule out prior processing. It is also noted that authenticity signals may disappear if the file has been modified, converted, or compressed. Anthropic specifies that if the C2PA standard expands to other formats, the tool will likely support them.

Verification via C2PA, Without Reading the File

Anthropic designed this tool to verify the use of Claude through the C2PA standard. The service is accessible at claude.com/check-content and allows users to upload a file for quick verification. The web interface is free, offering drag-and-drop functionality as well as the option to add files from the device or Google Drive on mobile. According to the site, the file remains on the device or storage service, with the tool only reading authenticity information and not storing or using the file for other purposes. If authenticity metadata is detected, the page confirms that the file shows signs of processing by Claude.

Texts Processed via a Private API and Discreet Watermarks Announced

The verification of texts generated by Claude is conducted through a separate Detection API, currently in private preview and reserved for eligible organizations, in accordance with European requirements. Anthropic plans to provide more information about this API in the coming months. Additionally, a more general AI detection tool is in development but is not yet available. A few weeks before the launch of the verification page, Anthropic announced the integration of digital watermarks in texts produced by Claude, stating that these watermarks do not affect quality, speed, or cost, that they are invisible and do not alter the appearance of the text, and that they remain attached even after copying and pasting. A registration form is available for organizations wishing to access the watermark detector. In reported tests, the tool was able to identify an image from a PowerPoint presentation converted in Claude. The verification page is now live.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.