Europe is asking whether its copyright framework needs another AI intervention
The European Commission has opened a targeted consultation on technology and copyright. One of its four specific areas is the use of copyright-protected content in artificial intelligence. The Commission says technological and market developments raise the question of whether the existing EU copyright framework should be reinforced through additional measures. Copyright and related-rights holders, generative-AI providers, other AI-value-chain participants and intermediaries are among those invited to respond. The consultation closes on November 3, 2026.
Why visual creators should care
The EU already has text-and-data-mining rules and AI Act obligations affecting general-purpose AI providers. Opening a new copyright consultation means the Commission is explicitly considering whether that framework is enough as generative-AI markets develop.
Practical takeaway
European photographers, illustrators, agencies and representative organisations now have a formal policy window in which to provide evidence about training, licensing, opt-outs, remuneration and enforcement rather than waiting for the next legislative proposal.
What to watch next
Watch the submissions before November 3 and, more importantly, whether the Commission subsequently proposes guidance, licensing infrastructure or legislative changes specifically addressing copyrighted works used by AI systems.
Sources
Stability’s new terms make source-image rights and provenance explicit user responsibilities
Stability AI’s revised Terms of Service became effective on September 30. Users who submit photographs, video or other material must represent that they hold the rights, licences and permissions necessary to provide those inputs. Stability says users retain their rights in submitted content and, to the extent permitted by law, Stability assigns its rights in generated outputs to the user. Stability may use content to improve its services and models, but its terms point users to an opt-out for preventing inputs and outputs from being used for model training. The terms also prohibit users from knowingly removing or circumventing Stability-applied watermarks, Content Credentials or other provenance markings identifying AI-generated or manipulated material.
Why visual creators should care
These are exactly the rights layers professional visual creators need to separate: permission to upload a source photograph, the provider’s licence to process it, possible model-training use, rights in the resulting output and the provenance attached to that output.
Practical takeaway
Before uploading client or third-party photography, establish that the project licence permits AI processing. Record the training preference that applied and preserve provider-applied provenance when the finished asset leaves the service.
What to watch next
Watch how Stability implements the training opt-out across its different products and third-party models, and whether Content Credentials remain intact when outputs move into common professional editing and publishing systems.
Sources
Training a LoRA creates a second rights chain beyond the generated image
Stability AI’s new terms separately address LoRAs and similar supplementary models. Data uploaded or incorporated into a LoRA is treated as user content, and the trainer is responsible for ensuring that the dataset, the LoRA and resulting outputs comply with applicable rights and policy. If a creator chooses to share a LoRA through Stability’s services, the terms grant Stability a licence to host and make it available and grant other users a licence to use the LoRA and create derivative works from its output.
Why visual creators should care
Custom visual models are increasingly something photographers and artists can build themselves. That does not remove the training-data problem; it moves responsibility closer to the individual studio.
Practical takeaway
Maintain a dataset manifest for custom models. Record every source image, creator or photographer, subject permission where relevant, licence basis and any restriction on machine-learning use before the image enters a LoRA dataset.
What to watch next
Watch whether major visual-AI providers standardise dataset documentation and whether commercial clients begin asking studios to provide evidence of the material used to train custom visual models.
Sources
A Delhi court is looking beyond deepfake generation to the platform tools that help reshape existing video
The Delhi High Court has directed social platforms to remove fake profiles, channels and deepfake videos allegedly exploiting educator Vikas Divyakirti’s identity. Reporting on the September 29 hearing says the court also questioned Instagram functionality that allows users to create new reels from another person’s video. The court left open whether editing or re-captioning lecture footage so that it changes the apparent meaning could infringe copyright or personality rights. Meta argued that intermediaries cannot be expected to monitor every misuse of platform features.
Why visual creators should care
Visual-rights disputes are beginning to reach beyond the initial generator. A platform can facilitate transformation and redistribution of authentic footage even when the resulting misuse is not a wholly synthetic video.
Practical takeaway
For creators publishing reusable video, copyright, likeness and contextual integrity can become separate enforcement routes. Keep originals and publication records so that altered versions can be compared with the source.
What to watch next
Watch the court’s treatment of platform remix tools and whether it develops a distinction between legitimate reuse and edits that materially misrepresent the person depicted.
Sources
AI-video buyers are starting to ask about provenance before vendors have finished building it
G2 analysed more than 10,900 verified reviews of AI video generators and surveyed three vendors about compliance. Of 43 product capabilities G2 tracks for the category, it says only watermarking directly addresses content origin or authenticity. Two of the three surveyed vendors reported growing customer demand for provenance systems such as C2PA or SynthID, AI-content disclosure, likeness and identity rights and content-safety controls, particularly from regulated industries and large brands.
Why visual creators should care
Compliance is becoming a procurement question for professional AI video rather than merely an ethical discussion. Agencies and brands increasingly need to know not only what a model can generate but what evidence and controls travel with the result.
Practical takeaway
When evaluating an AI-video supplier, ask about provenance, disclosure, likeness controls, source-image retention and export records alongside resolution, speed and price.
What to watch next
Watch whether provenance and consent management become formal product-comparison features rather than matters handled through support conversations and individual enterprise contracts.
Sources
A genuine photograph can make a fake image more believable
An AI-generated image falsely depicting College Park, Georgia mayor Bianca Motley Broom kissing a man circulated widely on Facebook. The Guardian reports that thousands saw the fabrication before it was debunked. Reporting around the incident says the synthetic image circulated in a context that also included a genuine photograph of the mayor with the same man, giving the false depiction an apparent connection to a real event. The mayor described the image as an attack that reinforced stereotypes about women in leadership.
Why visual creators should care
The provenance problem is not always a simple real-versus-fake comparison. Authentic photographs can be used as supporting material for a synthetic narrative, giving fabricated imagery contextual credibility it would not otherwise have.
Practical takeaway
When verifying a suspicious visual claim, do not stop after finding a genuine related photograph. Establish whether the image being circulated is actually that photograph or a synthetic reconstruction built around the same people or event.
What to watch next
Watch whether political campaigns and platforms adopt faster provenance checks for synthetic imagery and whether state lawmakers expand remedies for deceptive likeness use.
Sources
One photography competition is treating AI generation as a category question rather than a universal rule
The 2026 LumiTolia Photo Awards closed submissions on September 30 under rules that permit AI generation and broad digital manipulation only within the Conceptual/Fine Art category. Entrants must declare AI use. AI use in other categories results in disqualification. The rules also state that entrants retain copyright while granting the awards a limited licence for promotion, catalogues, exhibitions, workshops and related activities.
Why visual creators should care
Photography competitions are increasingly having to distinguish between documentary expectations and image-making categories where generative construction may be part of the artistic method. A single label saying AI is either allowed or banned often fails to describe that distinction.
Practical takeaway
Photographers entering competitions should check the rule for the specific category, not merely the competition’s general AI position. Keep enough production information to disclose what was generated or manipulated if asked.
What to watch next
Watch whether more photography competitions adopt category-specific AI rules and whether disclosure evolves into a standard field alongside conventional manipulation declarations.
Sources
