What changed: guidance became mandates
For two years the pattern was voluntary: industry bodies published best practice, platforms asked nicely, and disclosure was mostly a norms question. That has shifted to binding requirements in several markets.
- Labelling moved from recommended to required for AI-generated broadcast and streaming content in multiple jurisdictions.
- Human review at key production stages became a compliance obligation, not a quality preference.
- Platform-level disclosure rules increasingly sit alongside legal requirements, and are often stricter than the law.
The direction of travel is one-way: no major market has relaxed AI disclosure requirements. Build your workflow assuming disclosure is permanent and will tighten.
Who the rules apply to
Obligations generally attach to the party that produces or distributes the content, which in practice captures more people than expected.
| Party | Typical obligation |
|---|---|
| Broadcasters and streaming services | Label AI-generated programmes; maintain human review records |
| Production companies | Disclose AI use to commissioning platforms; keep review trail |
| Agencies and brands | Meet platform disclosure rules; substantiate claims in generated creative |
| Individual creators | Platform disclosure requirements; usually not statutory obligations directly |
If you supply footage into someone else’s regulated pipeline, expect to be asked to declare AI use contractually even where no law binds you directly.
What has to be labelled
The general principle is that synthetic content that could be mistaken for real footage must be identified. In practice that is broader than fully generated video.
- Fully AI-generated footage — clearly in scope everywhere.
- Substantially altered footage, including AI re-animation of real performances or synthetic replacement of a person’s likeness or voice.
- Synthetic presenters and cloned voices — usually treated as AI-generated even when the script is human-written.
- AI-assisted enhancement — a grey area. Colour grading and upscaling generally do not trigger disclosure; generative changes to content usually do.
The grey zone to get advice on: partial generation. If you shot a scene and generated one element into it, disclosure expectations are unsettled and vary by platform. Document your reasoning either way.
Where human review is mandatory
Several regimes now require a human to review at defined stages rather than merely at the end. This is a compliance requirement with an audit trail, not a creative preference.
- Review at key production stages, not only at final delivery.
- Final-cut review by a named human before publication.
- Prohibition on fully automated publication of AI-generated content in regulated contexts.
- Records sufficient to demonstrate review happened if asked.
If your workflow is fully autonomous — generate, publish, iterate — this requirement is the one most likely to force a process change.
Platform rules versus law
Two separate obligations apply, and satisfying one does not satisfy the other.
| Source | Scope | Enforcement |
|---|---|---|
| Law and regulation | Territory-specific, applies to regulated media | Regulator, fines or licence consequences |
| Platform policy | Applies to everyone on the platform | Demonetisation, removal, account action |
| Industry standards | Voluntary technical norms | Adoption pressure, interoperability |
For most working creators, platform policy is the binding constraint in practice because it is enforced automatically and immediately. Legal obligations matter more for broadcast, streaming and regulated advertising.
How labelling is implemented technically
Visible on-screen labels are the minimum. The more durable approach is machine-readable provenance attached to the file itself.
- Visible disclosure: an on-screen or on-player label the audience can see.
- Embedded metadata: provenance information carried in the file, resilient to re-upload.
- Content credentials: cryptographic provenance standards that record what was generated and by which tool, surviving screenshots and re-encoding less reliably but far better than nothing.
- Platform-declared disclosure: the checkbox or toggle when you upload.
Prefer embedded provenance: a visible label helps audiences now. Embedded provenance travels with the file and is what downstream platforms and regulators will increasingly check.
A compliance checklist
-
Map your markets
List every territory you publish into. Requirements differ, and the strictest applicable rule is the one to design against.
-
Classify your content
Decide in advance which of your outputs are fully generated, partially altered or merely enhanced, and write the classification down.
-
Add a human review gate
Name a reviewer for final cut. Record the review. Fully automated publication is not defensible in regulated contexts.
-
Turn on provenance
Use tools that embed content credentials, and keep provenance intact through your edit pipeline.
-
Declare at upload
Use platform disclosure controls even where not legally required. Non-disclosure that is later discovered is far more damaging than disclosure.
-
Re-check quarterly
This area changes fast. Assign someone to review current rules on a schedule.