01. The Real Risk Model
None of these show up in a quality review. They show up months later as a legal or reputational problem, which is why they need to be handled at generation time, not discovered retroactively.
02. Specific Risks
- Likeness and voice consent gaps: synthetic presenters or voices built from a real person's material without explicit, scoped consent for that use.
- Source licensing gaps: reference material used for generation without a clear licensing basis, especially for stock or third-party assets.
- Disclosure failures: synthetic content presented without appropriate labeling where jurisdiction or platform policy requires it.
- Provenance gaps: no record of what generated an asset, making later rights or disclosure questions unanswerable.
- Retention and deletion gaps: no policy for how long generated assets, source references, or consent records are kept, or how they're removed if consent is later withdrawn.
A common misconception is that a general employment agreement or a signed model release from years ago covers any future synthetic use. It usually doesn't. Consent scoped to a specific campaign, a specific use case, and a specific duration is what actually holds up, and a governed pipeline treats broad or expired releases the same as no consent at all until someone confirms otherwise.
03. Defenses That Work
- Synthetic presenters built only on explicitly licensed or consented sources, documented per use case.
- A provenance record for every asset: models used, source references, licensing basis, consent basis, approver.
- A deliberate disclosure policy applied consistently, not decided ad hoc per project.
- Legal or compliance sign-off on any real likeness use, before generation, not after delivery.
- A retention and deletion schedule for generated assets and their consent records, reviewed on a fixed cadence rather than left open-ended.
04. What to Ask a Vendor
- How is consent for real likeness or voice documented and verified before generation?
- What's logged for provenance, and can a specific asset's sourcing be reconstructed later?
- What's your default disclosure policy for synthetic content, and is it configurable to ours?
- What retention period applies to consent records and source references, and how are they purged on request?
See our AI video generation guide for how these controls fit into the full pipeline.
05. Frequently Asked
Can we use a real employee's likeness in a synthetic presenter?
Only with explicit, documented consent covering the specific use case, since likeness rights and consent scope vary by jurisdiction and can't be assumed from general employment terms. A properly governed pipeline treats this as a legal sign-off, not a technical step.
Do we need to disclose when video is AI-generated?
Increasingly yes, depending on jurisdiction and context, especially for synthetic likeness or voice. Disclosure policy should be decided deliberately and applied consistently, not left to default platform behavior.
What happens if a reference asset we generated from turns out to have licensing issues?
This is exactly what provenance tracking is for. Logging source references and their licensing basis at generation time makes it possible to identify and remediate affected assets quickly rather than discovering the scope of exposure after the fact.
Who should sign off on disclosure decisions, legal or marketing?
Legal or compliance should set the policy, since disclosure requirements are jurisdiction-dependent and carry real regulatory exposure. Marketing can execute within that policy, but the policy itself shouldn't be decided ad hoc by whoever happens to be producing a given asset.
Cloudz Computing builds synthetic presenters only on explicitly licensed or consented sources, with disclosure decided deliberately, not by default.
Explore the AI Video Generation solution →
Request a private consultation