How to Vet an AI Production Studio for Brand Safety: A Checklist
Vetting an AI production studio for brand safety means asking for evidence in six areas before you sign: where its models and source material come from, whether every rights holder in the chain consented in writing, how your files and any personal data are handled, what records exist for each delivered asset, who is liable if an asset infringes, and how outputs are marked and disclosed. A sound studio answers each question in writing and with a document, not with a reassurance. A studio that cannot show a rights chain, a data processing contract and per-asset records is asking you to carry risks it has not measured. The checklist below turns the six areas into concrete questions and describes what a good answer looks like. It is a procurement aid, not legal advice; contract terms and regulatory duties differ by market and should be confirmed with counsel.
Why brand safety is a different question from quality
Looking at a studio's portfolio tells you what it can make. It does not tell you whether you can use what it makes. Brand safety sits behind the image: the rights that allow publication, the handling of your confidential material, the proof that a human directed the work, and the allocation of blame when something goes wrong.
That makes vetting a different exercise from choosing a partner on creative merit. The question here is narrower: what could this production relationship expose the brand to, and can the studio show how it controls that exposure?
1. Source material and models
Ask what the studio generates with, and what that material was built from.
Which generation systems does the studio use, and under what licence terms? Commercial use of outputs should be expressly permitted by the system's terms. A studio that cannot name its systems or show the terms cannot show you the right to deliver.
Is any model trained or fine-tuned on your material? If yes, ask where the trained model lives, who can access it and what happens to it when the project ends.
Is any model trained on a real person's likeness? If yes, the consent has to be specific to that use. Model Likeness Rights in AI Campaigns lists what that consent has to say.
What happens to material you supply? Reference images, product files and brand guidelines should not be reused for other clients or fed into systems that retain them for their own training.
2. The rights chain
Every element in a delivered image has an origin, and each origin needs a permission that covers your use. Ask the studio to map it:
models and talent, with the scope of their consent: media, territories, duration and the digital replica specifically
photographers and retouchers, with assignment or licence terms
stock, textures, fonts and other third-party elements, with licence terms that allow use in advertising
your own assets, which you should be able to verify you are entitled to supply
The chain breaks at its weakest link: a signed model release does not help if a background element carries a licence that excludes advertising use.
3. Handling of your data and personal data
Studios receive unreleased product, pricing and campaign plans, and often images of real people. Both need protection.
Confidentiality: ask who has access to your files, how they are stored and when they are deleted.
Sub-contractors: ask which freelancers or other companies touch the work. Your approval should be required before new ones are added.
Personal data: where the studio processes personal data on your behalf, EU law expects a contract. Article 28 of the GDPR requires a controller to use only processors that provide sufficient guarantees, and requires a binding contract under which the processor acts only on documented instructions, keeps the data confidential, uses further processors only with authorisation, and deletes or returns the data at the end of the service at the controller's choice.
4. Records for each asset
For every delivered asset, a studio should be able to say how it was made. This is not bureaucracy; it is the evidence you need if a rights holder, a retailer or a regulator asks.
Ask what the file record contains:
the systems and versions involved
the source and reference files used
who directed and edited the work, and what they changed
the approvals given and by whom
the licence status of every third-party element
The record also matters for protection of your own work. In its January 2025 report on copyrightability, the US Copyright Office concluded that copyright does not extend to purely AI-generated material and that prompts alone do not provide sufficient control, while human authorship in creative selection, arrangement and modification of outputs can be protected. How much human contribution suffices is assessed case by case. A studio that documents human direction and editing gives you something to point to; a studio that does not leaves the question open.
5. Liability and remedies
Contract terms show how a studio really sees its risk. Ask:
Warranty: does the studio warrant that it holds the rights to deliver the assets for the agreed use?
Indemnity: will it cover claims arising from its own sourcing or processing, and are those limits stated?
Insurance: does it carry professional or media liability cover, and can it show a certificate?
Takedown: what is the process and the timeline if an asset has to be withdrawn or replaced?
Retention: how long will it keep source files and records so claims can be answered after the campaign ends?
Expect some limits; no studio accepts unlimited liability. What you want to see is that the limits are stated and proportionate to the campaign's value, not hidden in standard terms.
6. Marking, disclosure and review
Two questions belong here. The first is technical: how are outputs marked? Article 50(2) of the EU AI Act requires providers of image-generating systems to ensure outputs are marked in a machine-readable format and detectable as artificially generated. Ask which systems the studio uses, how marking is applied, and whether it survives retouching and export.
The second is organisational: who discloses to the audience, and who approves what? The wording and placement of a label, and the party responsible, belong in the contract. EU AI Act Labelling Rules for AI Ad Imagery sets out the deployer duty in detail.
Finally, ask who reviews the work for brand fit and accuracy: hands, fabric behaviour, product details, cultural fit. Where that review sits and who signs it off should be a named role, as described in How AI Production Studios Work With Ad Agencies.
Reading the answers
Studios that have done this work produce templates, records and contract terms on request, name their systems and roles, and state what they will not warrant. Treat a refusal to answer in writing as an answer, and attach the answers to the contract rather than leaving them in a pitch deck. For the wider regulatory picture, see Legal Framework for AI Campaigns.
Frequently Asked Questions
What should I ask an AI production studio before hiring it?
Ask six things: which systems and source material it uses, how the rights chain for models and elements is documented, how your data and personal data are handled, what record exists for each asset, who carries liability if an asset infringes, and how outputs are marked and disclosed. Get the answers in writing and attach them to the contract.
Who owns an AI-generated campaign image?
It depends on the contract and the jurisdiction. In the United States, the Copyright Office concluded in January 2025 that purely AI-generated material is not protected and that prompts alone do not give sufficient human control, while human selection, arrangement and modification can be. Agree ownership and usage rights in writing, and ask the studio to document human contributions.
Does the studio need a data processing agreement with us?
If the studio processes personal data on your behalf, for example images of identifiable people, the GDPR requires a contract that meets Article 28. That contract must bind the studio to act on documented instructions, keep data confidential, use sub-processors only with authorisation and delete or return data at the end.
Do AI-generated ad images have to be marked?
Providers of image-generating AI systems must ensure outputs are marked in a machine-readable format under Article 50(2) of the EU AI Act. Separate disclosure duties for those who publish deep fake content apply to deployers. Ask the studio how marking is applied and who handles disclosure.
What contract terms matter most for brand safety?
Warranty that the studio holds the rights to deliver, indemnity for claims arising from its own sourcing, stated limits of liability, a takedown process with timelines, approval of sub-contractors and a record-retention period. Limits are normal; what matters is that they are stated and proportionate.
About The Bridges AI
The Bridges AI is an independent AI and hybrid campaign production company working in strategic partnership with Most Wanted Models®, the German model agency founded in 1997. Combining nearly 30 years of fashion, talent and campaign expertise with generative AI, The Bridges AI produces high-end imagery and video for fashion, beauty and premium brands worldwide.
Written by the editorial team of The Bridges AI. Last updated October 2026.
Comments