Meta’s automated ad tools are altering photographs, products and approved campaign messages
Advertisers and agency executives say Meta’s AI creative tools have produced misleading or unusable variations of approved advertising assets. Reported examples include changing a photographed pajama dress into a shirt and trousers, adding a man to advertising for a women’s organisation, distorting product text and generating a bicycle with two handlebars. Some advertisers also reported bugs that activated AI settings they believed had been disabled.
Why visual creators should care
The platform is no longer acting only as a distributor. Its systems can become an unapproved retoucher, designer and image generator after the creator and client have signed off on the campaign. That creates risks around product accuracy, model presentation, advertising claims, creator reputation and responsibility for the final published image.
Practical takeaway
Agencies should preserve the approved master files and take dated screenshots of every automated creative setting before launch. Campaign agreements should state whether the media platform may crop, extend, generate, replace or materially alter visual assets, and who must review every resulting variation.
What to watch next
Watch whether Meta provides reliable account-level controls, complete change logs, stronger approval gates, compensation for damaged campaigns, and a dashboard showing every AI-generated variation before publication.
Sources
Australia’s AI investment debate is becoming a copyright-licensing fight
Australian government documents show that Anthropic raised uncertainty over copyright liability as a concern for investment. Officials recorded the company’s position that licensing from the long tail of smaller rights holders can be difficult, while government advisers urged engagement with rights holders so creative and media industries are fairly compensated for material used in AI training.
Why visual creators should care
Photographers, illustrators, filmmakers and independent visual artists make up much of that long tail. Treating the number of smaller rights holders as an obstacle could create pressure for broader training exceptions or collective solutions that reduce individual control over whether work enters commercial datasets.
Practical takeaway
Australian creators and organisations licensing work into Australia should keep ownership, publication, metadata and licensing records in order. Industry submissions should distinguish the administrative difficulty of obtaining permission from the legal question of whether permission and compensation are required.
What to watch next
Watch the Australian government’s forthcoming copyright position, any proposal for a text-and-data-mining exception, collective licensing or remuneration scheme, and whether creators receive meaningful control rather than compensation alone.
Sources
A synthetic campaign video is testing whether political disclosures are enough
A Michigan political advertisement used an AI-generated version of the state attorney general and included a statement that the video depicted speech and conduct that did not occur. The campaign illustrates how Michigan’s political-deepfake law functions in practice while prompting questions about whether disclosure neutralises the persuasive or reputational effect of a convincing synthetic performance.
Why visual creators should care
Visual creators and political-production teams may be asked to build synthetic campaign scenes that remain legally risky even when labelled. A disclosure does not necessarily resolve separate questions involving likeness rights, defamation, false endorsement, platform rules or the use of protected source footage.
Practical takeaway
Campaign producers should document the source images, footage, voices and models used to construct a synthetic person. Review disclosure placement and duration before release, and obtain legal approval separately from ordinary creative approval.
What to watch next
Watch whether Michigan receives further campaign-finance complaints, how regulators evaluate the prominence of disclosures, and whether lawmakers strengthen restrictions before the November elections.
Sources
AI-generated science images are making genuine photography harder to trust
A visual-science researcher warns that synthetic images are entering scientific illustration, research communication, education and laboratory workflows while becoming increasingly difficult to distinguish from genuine photographs and research figures. The concern is not only that fabricated images may be accepted, but that authentic photographs may also be dismissed as synthetic.
Why visual creators should care
Scientific and documentary photography derives much of its value from its connection to an observed event. When journals and audiences cannot reliably separate capture, enhancement, illustration and fabrication, photographers may be required to provide stronger provenance and production records with every image.
Practical takeaway
Scientific photographers, institutions and publishers should label generated illustrations separately from documentary photographs, retain raw captures and edit histories, and require disclosure of generative fill, compositing, synthetic data and material image manipulation during submission.
What to watch next
Watch whether research journals strengthen image-integrity checks, require Content Credentials or raw files, and create clearer distinctions between illustrative AI imagery and evidence used to support scientific findings.
Sources
Vehicle-image disputes show why photographers need proof tied to exact training files
Recent analysis of lawsuits involving Evox Productions describes claims concerning commercial vehicle photographs allegedly used in AI training. The disputes highlight a recurring litigation problem: broad evidence that a model was trained on web images may be less useful than records connecting identifiable copyrighted files to a particular dataset, download process or model pipeline.
Why visual creators should care
Commercial photographers often produce highly structured image libraries with consistent angles, naming systems and licensing histories. Those records may make it easier to identify exact works and demonstrate systematic copying than it is for creators whose files have circulated without persistent identifiers.
Practical takeaway
Maintain original filenames, embedded metadata, checksums, registration records, licensing histories and delivery logs. Where practical, keep a searchable catalogue that connects every published derivative to its high-resolution master and copyright record.
What to watch next
Watch whether the claims survive early procedural challenges, whether dataset records identify particular Evox files, and whether courts treat systematic ingestion of a commercial photography catalogue differently from incidental web collection.
Sources
AI copyright cases are becoming fights over whether the evidence still exists
The New York Times, Daily News and other publishers have asked a federal judge to sanction OpenAI, alleging that relevant datasets and system logs were withheld or destroyed. OpenAI disputes the accusations and says restrictions on sharing logs protect user privacy. The court has not yet determined whether misconduct occurred.
Why visual creators should care
Visual artists pursuing training claims face the same fundamental evidentiary problem: they may need records held exclusively by the model provider to show whether a particular photograph, illustration or video entered a training dataset. If those records are incomplete or unavailable, proving copying becomes substantially harder.
Practical takeaway
Creators should preserve their side of the evidence before a dispute begins: publication dates, page captures, URLs, metadata, copyright registrations, suspected model outputs and any responses obtained from dataset-search or opt-out tools.
What to watch next
Watch the court’s ruling on sanctions, what dataset and log searches are considered technically possible, and whether future AI litigation orders require model companies to preserve training and output evidence earlier.
Sources
