Brazil has now defined what counts as an AI deepfake in election advertising
Brazil’s Superior Electoral Court has adopted a formal definition of deepfake for the 2026 elections. By a 5–2 vote, the court defined it as synthetic content produced or manipulated by AI or equivalent technology that has a degree of realism or plausibility and creates, reproduces or alters the image, voice or manifestation of a living, deceased or fictional person. The ruling arose after an AI-generated video recreated former president Jair Bolsonaro endorsing his son Flávio Bolsonaro. The court declined to penalise that particular use because it occurred during a party convention that it did not classify as electoral advertising.
Why visual creators should care
The court is separating two questions that are often collapsed together: whether a depiction is a deepfake and whether a particular use of that depiction is legally prohibited. For filmmakers, agencies and visual creators working around public figures, context remains important even when everyone agrees the image itself is synthetic.
Practical takeaway
Record both the nature of the synthetic depiction and its intended use. A realistic AI likeness can trigger different rules depending on whether it appears in advertising, journalism, satire, entertainment or another context.
What to watch next
Watch how the TSE applies its definition during Brazil’s October election campaign and whether platforms adopt the same realism-based definition when reviewing synthetic political imagery.
Sources
Adobe can now turn the context inside a Slack conversation into an image or video
Adobe launched Adobe for Slack on September 2 for Slack Business+ and Enterprise+ teams. The integration exposes more than 70 Adobe tools including Firefly, Photoshop, Lightroom, Premiere, Illustrator, Express and Adobe Stock through Slackbot. Users can ask Slackbot to draw context from conversations, shared files, channels and Canvases and then use Adobe tools to generate or edit images, videos and campaign assets. It can also retrieve existing Creative Cloud assets and create channel-specific variations.
Why visual creators should care
The source material entering an AI-assisted visual workflow is no longer necessarily a file intentionally uploaded to a generator. It can be the wider project context already sitting in collaboration software: briefs, references, campaign conversations and previous creative assets.
Practical takeaway
For confidential projects, treat AI access to collaboration context as part of the production permission. Teams should know which conversations, files and Creative Cloud assets may be pulled into a generative request rather than assuming Slack context is separate from the image-making system.
What to watch next
Watch what enterprise controls Adobe and Slack expose around which channels, conversations and assets an AI-assisted creative request may access and how generated derivatives retain provenance.
Sources
Anthropic has made Claude’s file provenance readable without sending the image back to Anthropic
Anthropic has launched a browser-based tool for checking whether supported files carry a C2PA Content Credential indicating that Claude created or processed them. The checker accepts common image formats including JPEG, PNG, TIFF, HEIC, AVIF, SVG and DNG, as well as video and audio. Anthropic says the check runs locally in the browser, the file never leaves the device and the credential contains no identifying information about the user.
Why visual creators should care
That makes provenance verification more usable for photographers, agencies and picture desks handling confidential material. A client can inspect the attached credential without uploading the entire asset to another verification service.
Practical takeaway
Treat a positive credential as evidence that Claude participated in the file’s production, not proof of authenticity, copyright ownership or clearance. Equally, a missing credential is not proof that AI was never involved because metadata can be lost during editing, screenshots or conversion.
What to watch next
Watch whether more generator providers adopt local C2PA checking and whether DAM and newsroom systems begin performing the same verification automatically at ingest.
Sources
Envato allows an AI image to advertise a product that is not itself allowed to contain AI-generated assets
Envato’s current policy for Market and Elements prohibits creators from selling AI-generated material as downloadable content or as the primary component of an item. That restriction applies across content types. AI-generated imagery can, however, be used in item previews where the creator has the appropriate commercial licence and the generated asset is not included in what the customer downloads.
Why visual creators should care
The same AI image can be acceptable marketing material and unacceptable licensed content depending on where it sits in the product. For illustrators, designers and template creators, preview rights and end-customer asset rights are therefore separate checks.
Practical takeaway
When AI imagery appears in a product mock-up or preview, keep it clearly outside the downloadable customer package unless the marketplace explicitly permits otherwise. Also retain the generator licence that permitted the promotional use.
What to watch next
Watch whether Envato relaxes the rule as provenance and licensing systems develop, or instead begins requiring explicit AI disclosure even for preview-only imagery.
Sources
A face-swap platform is making consent something the user has to confirm before generation
PixelDojo updated its creator and developer rules at the start of September. Its restored Face Swap workflow now requires users to confirm that the adults shown in both source photographs agreed to participate, and uploaded images pass safety screening. Its new Creator Agreement separately requires users to confirm that they hold rights to uploaded material and have consent from depicted people. A September 2 Terms update also warns API and MCP developers that individual underlying AI models may carry their own usage rules.
Why visual creators should care
Consent is moving from a paragraph in the general terms into the moment the synthetic image is actually created. The model-specific rule is important too: an aggregator may provide a single interface while different underlying generators carry different restrictions.
Practical takeaway
For face swaps or synthetic talent work, retain the underlying consent rather than relying only on clicking the platform confirmation. When using an aggregator through an API, record the actual model used and check its provider-specific restrictions.
What to watch next
Watch whether visual platforms begin attaching consent records and underlying-model identifiers to provenance metadata rather than keeping them only in account logs.
Sources
Commercial-use permission still does not mean the generator cleared the image for you
Donatello’s current terms allow outputs from its image and motion production routes to be used commercially for client work, advertising, websites and products. At the same time, the service says that commercial-use permission does not grant rights in third-party copyrighted material, trademarks, privacy, publicity or likeness contained in an input or output, and does not guarantee that output is unique, non-infringing or copyrightable. Its terms also say commercial-use permission does not include using the output to train or fine-tune another AI model without separate permission.
Why visual creators should care
A platform can legitimately permit commercial exploitation of what it generates without being able to clear every protected element that happens to appear in the result. Those are two different legal questions.
Practical takeaway
Treat 'commercial use allowed' as permission from the platform, not a rights-clearance certificate. Before client delivery, review recognisable people, logos, characters, artworks and other third-party material separately.
What to watch next
Watch whether smaller visual generators begin offering human review or indemnification layers rather than leaving all third-party rights assessment with the creator.
Sources
