AFP wants provenance to travel with news video from the production system
Agence France-Presse and Dalet announced a collaboration on September 25 to integrate C2PA Content Credentials into video and encoding workflows. Engineers from the two organisations will develop a proof of concept showing how a news agency can attach verifiable origin and history information to video as it is prepared for onward delivery to broadcasters. The PoC is expected to be completed in the first quarter of 2027. AFP argues that both AI and non-AI material can benefit from provenance information and describes C2PA as a way of attaching a verifiable certificate to real field journalism.
Why visual creators should care
For professional photographers and filmmakers, provenance becomes much more useful when it survives the production and distribution chain. If credentials are attached only after publication, there are more opportunities for the connection between source and finished asset to be lost.
Practical takeaway
For editorial and documentary workflows, think about provenance at ingest and export rather than treating it as a final publishing label. The useful record should travel with the asset through editing, encoding and delivery.
What to watch next
Watch the AFP/Dalet proof of concept in early 2027, particularly whether broadcasters can preserve and expose the credentials after their own transcoding and distribution processes.
Sources
YouTube is beginning to treat a creator's voice as part of their detectable AI likeness
YouTube's Made on YouTube 2026 announcement expands its synthetic-likeness protections. The platform says billions of channels can sign up for automated deepfake detection and use its likeness detection tool to identify AI-generated content depicting their visual likeness. Later this year, YouTube plans to integrate speaking-voice detection with facial detection to improve match accuracy and lay the groundwork for broader voice protections. It also plans to bring likeness review and action tools into the YouTube mobile app.
Why visual creators should care
A synthetic representation of a person is no longer only a face problem. For filmmakers, performers and creators, a recognisable voice can be part of the same digital replica even when the visual and audio components are generated separately.
Practical takeaway
When obtaining permission for synthetic presenters or performers, distinguish face, voice and combined digital-replica uses. Permission to use a portrait does not automatically establish permission to clone or synthesise the person's voice.
What to watch next
Watch how YouTube handles false matches, whether voice-only synthetic content becomes independently searchable and what evidence creators must provide when requesting removal.
Sources
A real photograph can now trigger a biometric identity check before an AI video model will use it
Picacho updated its privacy policy on September 25. The AI-character platform says some video models refuse photographs of real people unless the person verifies that the likeness is their own. For supported Seedance workflows, users can undergo an optional live face check operated by BytePlus. BytePlus keeps a reference image after successful verification and can compare up to three selected character photographs against it. Picacho says the process involves images, live-check video and extracted facial features, requires explicit consent, and is available only for verifying the user's own face. Facial information is stored by BytePlus in Southeast Asia until verification is removed or the account is deleted.
Why visual creators should care
Safety controls around AI likeness generation can themselves create a biometric-data workflow. Preventing unauthorised use of a face may require processing and temporarily retaining highly sensitive information about that same face.
Practical takeaway
If a production tool requires identity verification, document who performs the verification, what facial information is retained, where it is stored, how long it remains and how consent can be withdrawn.
What to watch next
Watch whether identity verification becomes standard across AI video models that permit real-person reference images and whether interoperable verification can avoid repeating biometric checks across multiple providers.
Sources
Stanford replaced a real student inside a photograph and then removed the result
Stanford University has removed campus promotional banners after confirming that AI was used to alter an existing photograph of three students. Reporting says the original image was made at a university dining hall. In the altered version, Latino student Billy Ramirez was replaced with an AI-generated Black woman, while the appearances and facial features of the remaining students were also modified. Stanford said both the AI alteration and the failure to disclose the use of AI violated university policy and removed the banners.
Why visual creators should care
This is a particularly clear example of the difference between editing a commissioned or institutional photograph and changing who the photograph appears to document. Once AI replaces or materially reconstructs identifiable people, the image can stop functioning as a straightforward record of the event it originally captured.
Practical takeaway
For documentary, institutional and promotional photography, define a boundary between corrective editing and generative alteration of the people depicted. If a person is replaced or materially reconstructed, disclosure should not depend on whether a viewer notices.
What to watch next
Watch whether universities, agencies and corporate communications teams adopt explicit rules governing generative alteration of identifiable people in documentary-style promotional photography.
Sources
AI filmmaking in China is becoming an industry before the copyright rules are settled
Reuters reports that Chinese cities and industrial parks are competing to attract AI filmmakers and studios with subsidised rent, computing support and other incentives. Production costs for AI short dramas have fallen sharply, and hundreds of thousands of AI shows were launched on Douyin during the first half of 2026. China's National Film Administration has also granted a public-screening licence to the 90-minute science-fiction film Sanxingdui: Future Memories, described as the first AI film produced by a major Chinese studio to receive approval for theatrical release. The rapidly expanding industry is still waiting for clearer copyright rules.
Why visual creators should care
AI filmmaking is moving from experimental clips into professional studios, distribution and theatrical release. That increases the commercial importance of unresolved questions around training material, generated assets, performers, reference imagery and ownership.
Practical takeaway
For AI-assisted film production, preserve the rights chain now rather than waiting for the law to settle. Record model licences, reference assets, human contributions, likeness permissions and generated elements while the production is happening.
What to watch next
Watch China's emerging copyright and production rules for AI film, particularly requirements around training data, disclosure and rights clearance for commercially distributed productions.
Sources
A court has ordered removal of a disputed AI face-swap built from a real woman's photograph
The Delhi High Court ordered Meta to remove disputed content within 24 hours in a case brought by a woman who alleges that her photograph was used without consent to create an AI-generated and morphed image depicting her alongside Indian Prime Minister Narendra Modi. The court also directed police to provide protection to the petitioner and issued notices to several respondents. The allegations remain before the court; no final finding on the origin or authenticity of the disputed image has been made.
Why visual creators should care
The case shows how quickly an ordinary source photograph can acquire an entirely different political, reputational and personal meaning after face-swapping. Rights in the original photograph and consent from the person depicted are separate issues from the legality of the synthetic representation.
Practical takeaway
Do not treat access to a person's photograph as permission to place that person into a new generated context. For synthetic likeness work, record the specific representation and intended context that the person agreed to.
What to watch next
Watch the next hearing and whether the court develops a more detailed approach to platform removal, source-image consent and responsibility for circulating AI-manipulated likenesses.
Sources
