September consultation update

Since our August update, we have continued the public consultation, started testing the HPF declaration across multiple productions, and expanded our research into how AI is being used across film, television, and animation. This work has informed v0.9.3 of the draft standard and helped narrow the questions that still need to be tested before v1.0.

What we heard

• A declaration needs a clearer relationship to the version of a work it covers. Later cuts and substantive localisations need to be distinguishable from technical deliveries of the same released version.
• Descriptors can overlap, but some combinations need clearer guidance. This is particularly relevant to AI use involving actors and recorded performances.
• AI use during production and AI provenance in a finished work are different questions. Research found many operational, intermediate, temporary, transformed, or discarded uses that are not represented simply by looking at the finished work.
• Producers may not always know where AI has been used. This is particularly relevant in post-production, where AI or machine-learning functions may be built into existing tools or may not be obvious from the finished work.

These findings come from consultation responses and direct feedback, alongside our wider research into production workflows. That research now covers more than 300 documented uses and attempted uses across development, editorial, VFX, animation, sound, localisation, technical work, and other areas of film and television production. It is not a representative sample of industry practice, but it provides a wider set of cases against which to test the draft standard. The declaration is also being tested across multiple productions, with producer follow-up planned as the next stage.

v0.9.3 updates based on consultation feedback and research

The September update does not change HPF's three classifications: No AI Used, Assistive AI, and Generative AI. Nothing in the research or consultation feedback so far has identified a need for an additional category.

More precise guidance now explains how declarations apply to different versions of a work. A declaration can travel with different deliveries of the released version it identifies, while a later director's cut, revised edit, or substantive localisation has its own declaration. We have also removed the previous requirement to notify recipients of an earlier declaration when a later one is made.

Descriptors can be combined where they describe different aspects of the same AI use. HPF will not introduce percentage thresholds, departmental weighting, or an overall AI score; descriptors and a short factual summary remain the current approach. The descriptors themselves remain provisional while testing continues.

New guidance covers temporary and reference material, transformed output, and restoration, including the distinction between reconstructing or processing supplied material, which is Assistive AI, and fabricating new content, which is Generative AI.

These decisions have narrowed the remaining consultation questions, which now focus on reasonable enquiry, AI use involving actors and recorded performances, difficult finished-work and version cases, and implementation. The full record of changes and consultation decisions is available in GOVERNANCE.md.

What the research is showing

The research examined what happened to AI contributions as they moved through production. Many documented uses were operational, intermediate, temporary, transformed, or discarded before delivery, while in a substantial number of cases the available evidence did not establish whether the contribution reached the finished work.

This supports HPF's continued focus on finished-work provenance rather than attempting to record every use of AI during production. An intermediate or operational use may affect production without appearing in the released work, while material that does reach the finished work may have undergone substantial human transformation first.

Testing HPF against these cases has identified questions around transformed output, temporary and intermediate material, AI operations that affect finished work without leaving a discrete generated asset, and the boundary between reconstruction and generation. The study does not measure how prevalent particular practices are or validate the taxonomy.

Testing the declaration with productions

The multi-production test is looking at whether producers can establish the information needed to complete a declaration and provide a useful factual account without documenting every tool or process used during production.

One question is what reasonable enquiry into AI use should involve, particularly in post-production. AI or machine-learning functions may be built into existing software, enabled by default, or difficult to identify from the finished work, so producers may need information from editors, post-production supervisors, VFX teams, or vendors.

The next stage will involve following up with participating producers about how they established the information in their declarations and where additional guidance may be needed.

If you are a producer and would like to test making an HPF declaration for a current or completed production, please contact contact@humanprovenance.film.

What we will test before v1.0

We are extending the provisional descriptor survey until 31 October. Responses so far have raised questions about AI use involving actors and recorded performances, including changes to recorded voice, expression, movement, or timing; uses involving several descriptors; and synthetic content that incorporates human performance.

As planned earlier in the consultation, pre-v1 working groups will run in November and December. We are recruiting additional participants, who will receive a short set of anonymised production scenarios and can join an online session of around 75 minutes or provide written feedback.

The working groups are expected to cover:

• alteration of actors' recorded performances, digital replicas, generated voice, and synthetic performances;
• mixed Assistive and Generative workflows;
• transformed output and intermediate VFX or animation cases;
• restoration and reconstruction;
• later versions, localisation, subtitles, captions, and episodic releases; and
• what information producers can reasonably establish when preparing a declaration.

The findings will help determine whether these cases require changes to the taxonomy, clearer guidance, additional examples, or no change before v1.0. We are particularly looking for producers and practitioners working in post-production, VFX, animation, sound, localisation, casting, and actor representation.

The public consultation will close on 31 October 2026 as planned. The November and December working groups will then focus on the questions that remain before v1.0. Participation does not imply endorsement or adoption of HPF, and interested practitioners can register through the consultation contact route.