India says the AI artwork can be original — but the AI cannot be its author
India’s Registrar of Copyrights has rejected Stephen Thaler’s application to register the visual artwork A Recent Entrance to Paradise with his DABUS AI system identified as author. The decision did not find the work itself insufficiently creative. The Registrar concluded that the image satisfied the originality threshold under India’s Copyright Act, while holding that DABUS could not occupy the statutory role of author because it is neither a natural nor legally recognised juristic person. On the facts before the Office, Thaler was identified as the person who caused the work to be created because he configured DABUS, supplied visual and linguistic inputs and initiated the process. The application nevertheless failed because Thaler declined an opportunity to amend it and name himself as author.
Why visual creators should care
For AI image creators, this is substantially more nuanced than the position that an AI-generated image is automatically uncopyrightable. India’s statutory framework expressly addresses computer-generated works by asking who caused the work to be created. The difficult question therefore shifts towards identifying the human creative and causal contribution.
Practical takeaway
If copyright in AI-assisted visual work matters, preserve evidence showing what the human actually contributed: source photographs, selection, configuration, prompts, curation, rejected outputs, compositing, editing and the decisions leading to the final image. Do not identify the AI tool itself as though it owns or authors the work.
What to watch next
Watch whether the ruling is challenged and how India applies the 'person who causes the work to be created' test to modern prompt-driven generators where human involvement is much lighter than Thaler’s described DABUS workflow.
Sources
China’s AI crackdown is now targeting synthetic impersonation at platform scale
Chinese cyberspace authorities say the latest phase of their campaign against misuse of AI has removed more than 5.61 million pieces of unlawful or rule-breaking content, taken action against more than 49,000 accounts and dealt with more than 2,400 websites and apps. The Cyberspace Administration of China says targeted conduct includes false information generated with AI, impersonation of real people and content affecting minors. Major platforms have strengthened systems for detecting manipulated faces and cloned voices and generators have been pushed to improve compliance with AI-content labelling requirements.
Why visual creators should care
China already requires generated images, video and other synthetic media to carry identifiers. This enforcement phase demonstrates that disclosure and likeness rules are not remaining theoretical platform obligations: they are becoming a basis for account, content and application enforcement.
Practical takeaway
Creators and agencies publishing synthetic people or materially generated visuals into Chinese platforms should treat AI labelling and likeness permission as delivery requirements, not optional metadata added after publication.
What to watch next
Watch whether regulators publish more granular enforcement examples involving commercial advertising, photographers’ source imagery and unauthorized use of identifiable faces.
Sources
Adobe is making C2PA provenance automatic rather than something the creator has to remember
Adobe’s current enterprise transparency documentation says supported generative-AI creation and editing workflows automatically receive C2PA metadata. The feature is enabled by default and cannot be disabled. The attached record can include the AI system and model used, whether the asset was generated or edited, timestamps and a unique identifier associated with an individual generative-AI use. Adobe says supported workflows preserve existing C2PA history when assets from several sources are combined, including supported non-Adobe credentials.
Why visual creators should care
For agencies, the provenance record is moving closer to becoming part of the production infrastructure rather than a voluntary label somebody has to add before publication. It can also contain much more than a binary statement that AI was used.
Practical takeaway
Preserve the provenance-bearing master and inspect what information is actually attached before client delivery. Visible disclosure and C2PA metadata are separate things, and downstream platforms may not display the embedded information even when it remains in the file.
What to watch next
Watch how consistently Adobe preserves detailed model and ingredient histories across Photoshop, Firefly, Workfront, AEM and third-party publishing systems.
Sources
ArtStation is turning an artist’s NoAI choice into something a scraper can read
ArtStation’s current AI guidance says projects tagged NoAI automatically receive an HTML NoAI meta tag explicitly signalling that the content must not be used by AI systems. ArtStation says absence of the tag is neither permission nor prohibition and leaves the question to applicable copyright law. Separately, generative-AI products distributed through the ArtStation Marketplace must be tagged CreatedWithAI and non-compliant products can be delisted.
Why visual creators should care
A human-readable sentence saying 'do not train on my work' may never be encountered by an automated collection system. ArtStation’s approach places that choice in machine-readable page metadata while keeping the underlying copyright position intact.
Practical takeaway
If you use ArtStation and do not want a portfolio project used for AI, apply the NoAI tag and retain evidence showing when it was added. The technical signal does not guarantee compliance by scrapers, but it creates a much clearer record of the artist’s expressed restriction.
What to watch next
Watch whether major dataset builders recognise ArtStation’s NoAI metadata and whether comparable signals converge around a common technical standard rather than platform-specific tags.
Sources
500px’s answer to an AI false positive is the photographer’s RAW file
500px’s current Licensing FAQ says AI-generated work is not eligible for its commercial licensing programme. It extends that rule to photography where more than 10 percent of the image has been edited with AI-powered tools. The platform also describes what happens when an authentic photograph is incorrectly classified as AI-generated: the photographer can challenge the flag, but 500px asks for the original RAW file as evidence of authentic capture.
Why visual creators should care
This is an unusually concrete example of authenticity evidence acquiring commercial value. A RAW file that once existed mainly for editing and archival purposes can now become the photographer’s proof that a photograph was actually captured rather than generated.
Practical takeaway
Keep RAW files for commercially important photography even after delivery. Where a stock platform or client restricts generated imagery, the original capture may be the simplest evidence available if an automated system gets the classification wrong.
What to watch next
Watch whether other stock libraries adopt measurable AI-edit thresholds and whether RAW verification develops into a broader camera-provenance or C2PA workflow.
Sources
AI is entering editorial photography through the caption even when the photograph itself is conventional
Current IMAGO image listings display an explicit notice that captions may have been translated or enriched by an AI system. The notice appears alongside conventional professional image information including photographer credit, recording date, media identifier, licensing terms and release information. A September 3 photograph of the Weimar Triangle foreign-ministers meeting carries the same AI-caption notice while remaining a conventionally credited editorial photograph.
Why visual creators should care
AI involvement in a picture workflow does not necessarily mean the pixels were generated. Automated translation or enrichment can alter the textual record that editors, search systems and buyers use to understand a real photograph, making factual metadata review part of AI governance too.
Practical takeaway
Distinguish AI-generated imagery from AI-generated information about imagery. For editorial work, protect the authoritative photographer caption, date, credit and rights fields and review automatically enriched descriptions before treating them as factual.
What to watch next
Watch whether agencies begin identifying exactly which caption fields were machine-generated and preserve the contributor’s original caption separately from subsequent AI enrichment.
Sources
