Adobe is making provenance survive beyond the image editor
Adobe updated its Coworker Campaigns documentation on September 26 with details of automatic C2PA handling for generative imagery. When a user generates or edits an image using generative AI in Coworker Campaigns, C2PA metadata is attached automatically without requiring the user to switch the feature on. Adobe says images subsequently sent through email campaigns retain that metadata, allowing recipients with compatible tools to inspect information about the image's origin and editing history.
Why visual creators should care
Provenance has limited professional value if it disappears as soon as the image enters the publishing system. Preserving the credential through campaign delivery moves Content Credentials closer to being infrastructure rather than an export option.
Practical takeaway
Do not test provenance only on the master file. Test the actual delivered asset after it passes through the client, CMS, email system or social platform that will publish it.
What to watch next
Watch whether Adobe extends the same automatic preservation through other campaign channels and whether receiving platforms expose the credential clearly enough for ordinary viewers to inspect.
Sources
Australia's AI copyright argument could reach photographs made anywhere in the world
New reporting on Australia's parliamentary AI inquiry explains a potentially global consequence of the copyright changes being discussed. Australian officials told the inquiry that copyright law generally applies according to the territory where the copying occurs. If model training takes place in Australia, an Australian exception could therefore govern copying of creative material originating elsewhere. Inside Story reports that one proposal under discussion would involve a compensation fund for participating Australian creators and an opt-out mechanism for creators unwilling to permit training. Overseas rightsholders would potentially need to know about and act on the Australian opt-out despite receiving no payment from the proposed Australian fund.
Why visual creators should care
A photographer in Switzerland, an illustrator in Britain or a filmmaker in the United States could potentially be affected by a training rule adopted in a country where they have never published or worked. The location of model training can therefore become part of the rights question.
Practical takeaway
Do not assume an AI-training rights dispute is relevant only to creators living in the jurisdiction changing its law. Track where the legally significant copying takes place and whether proposed opt-out systems expect foreign creators to register their works.
What to watch next
Watch the Australian parliamentary inquiry and whether any proposal survives scrutiny under the Berne Convention's restrictions on copyright formalities and its three-step test for exceptions.
Sources
Digitising an archive and training on the digitisation can be two different permissions
The Guardian reports that material digitised through Oxford University's partnership with OpenAI has been used to populate OpenAI training data. Internal university records obtained through freedom-of-information requests show that by June 2025, 125,000 scanned images from historical dissertations had been shared with OpenAI. Oxford says the material involved is out of copyright, the amount is modest relative to the Bodleian collection, the library retains rights in the scans and the digitised material will also be made publicly available.
Why visual creators should care
The story exposes a distinction that matters to museums, libraries, photographers and archives: copyright status of the underlying historical object, rights associated with a modern digitisation and permission to use that digitisation as training material are not necessarily the same question.
Practical takeaway
When licensing archive digitisation work, state explicitly whether machine-learning or model-training use is included. Do not rely on the age or public-domain status of the object being photographed to answer what may happen to the newly created digital file.
What to watch next
Watch whether cultural institutions begin writing AI-training permissions directly into digitisation contracts and whether photographers producing archival reproductions are told when those files may enter model datasets.
Sources
An AI comics platform is separating the right to sell an output from copyright in the output
GenToon has published revised terms for its AI webtoon platform, with the revision posted September 26 and scheduled to take effect October 3. The terms say rights to user-created content belong to users, while whether AI-generated work qualifies for copyright is determined by applicable law. Commercial use is allowed within the user's subscription licence. GenToon says prompts, uploaded reference images and generated outputs are not used to train its own or third-party AI models; generation uses the Google Gemini API. Users remain responsible for ensuring reference images do not infringe copyright, portrait, trademark or other third-party rights.
Why visual creators should care
Commercial permission and copyright ownership remain easy to confuse. A platform can contractually allow an illustrator or designer to sell a generated image without being able to guarantee that copyright subsists in that image.
Practical takeaway
Record the commercial licence and copyright position separately. If reference images were used, preserve the permission for those files as another separate part of the rights chain.
What to watch next
Watch whether visual AI services increasingly state explicitly that output commercialisation rights do not answer the separate legal question of copyrightability.
Sources
A major creative convention is drawing its AI boundary at generation rather than assistance
UK Games Expo has introduced a new policy prohibiting AI-generated games and merchandise from its show floor. Organisers reserve the right to require exhibitors to remove prohibited material and can close stands or ban exhibitors who refuse. The policy retains a narrower exception for AI used for minor editing, production assistance or accessibility.
Why visual creators should care
For illustrators and designers, this is a more useful rule than a blanket statement that 'AI is banned'. It attempts to distinguish a generated creative asset from software assistance used around a human-created work.
Practical takeaway
If work is entering a market, competition or client workflow with an AI restriction, record the actual AI operation. 'AI-assisted' is too broad to establish whether the work falls inside or outside a rule like this.
What to watch next
Watch how UK Games Expo interprets 'minor editing' when exhibitors use generative fill, upscaling, background reconstruction or other tools that sit between conventional editing and generation.
Sources
Some clients are beginning to reject AI-generated design before it reaches the audience
Venues and organisations in Bristol are introducing explicit restrictions on AI-generated promotional artwork. Café Kino has banned AI-generated posters, the University of Bristol Students' Union has prohibited AI artwork at its Freshers Week poster sale, and music venue Zed Alley has also stopped using AI-generated artwork on its social channels. The venues cite concerns including support for human designers, originality and audience response.
Why visual creators should care
AI policy is moving into ordinary commissioning decisions. For freelance illustrators and designers, whether generative imagery is acceptable can now be a client requirement rather than simply a personal production choice.
Practical takeaway
Ask about AI restrictions at the brief stage. If a client requires human-created artwork, agree what counts as permitted editing or production assistance before work begins rather than trying to establish compliance at delivery.
What to watch next
Watch whether agencies and venues begin putting no-generative-AI clauses directly into design briefs and supplier agreements, and whether they distinguish generative creation from tools such as masking, cleanup and upscaling.
Sources
