June consultation update

We are a month into the v0.9 consultation, which runs to 31 October. A short update on what people have told us, one change we have made, and where we still need help.

What we have heard

A few things keep coming up:

• Festivals want one definition they can point filmmakers to. They are asked about AI disclosure constantly and have nothing standard to hand over.
• Animation and VFX are the hard cases. Animation in particular seems to have the most generative AI use.
• Most of the AI use filmmakers describe is in pre-production: planning, previsualisation, scheduling. Much of it never reaches the screen, so the standard does not cover it. Getting the breadth of disclosure scope right is something we want to focus on in the coming rounds of feedback.
• Lawyers and practitioners raised harder questions at the edges: how to read a film that lands in the top category for a single AI shot, how a producer can honestly sign "No AI Used" when common software enables AI features by default, and how the standard sits alongside the EU AI Act.
• Insurance splits along clear lines. A specialist film insurance broker told us production cover is largely unaffected by AI use, and could even lower claims costs. The growing conversation is in errors and omissions cover, where what matters is how content such as scripts, music and images was generated.
• People value that it is free. Several of the organisations we spoke to said an open licence, free to use and adapt under CC BY 4.0, is part of why they would back it rather than a proprietary scheme.

People like the standard. The thing we hear most is that they want it to exist.

On signing up

At launch we invited organisations to join as signatories, and we spoke to several directly about it. During the consultation we have set that aside. Committing names before the standard is settled is premature. We are now looking at how organisations can join to support the standard's development and adoption, not just endorse it, and we will set out that model alongside v1.0.

One change: the question the taxonomy asks

We have changed the central question.

The old version was "enhance or replace": did AI enhance human work, or replace a human role? It caused more trouble than it solved. It asks you to guess whether a human would otherwise have done the job, and the word "replace" turns a disclosure question into an argument about labour.

The new version asks what is in the film:

• Is there any AI output in the finished film, as it goes out?
• If so, did the AI work from human material, or make new content itself?

That maps straight onto the three categories:

• No AI output: No AI Used
• AI worked from human material: Assistive AI
• AI made new content: Generative AI

The categories are the same. Now a producer answers facts they can put their name to, and the warranty carries the weight. Most of the detailed tests in the taxonomy already worked this way, so the main question has been changed to match them. The taxonomy page sets out the full reasoning.

We have also added a reconstruction test for tools that work on footage you already have. It separates AI that restores or repairs what a human captured, which is Assistive AI, from AI that fabricates content that was never there, which is Generative AI.

Where we want more feedback

The questions we would most like help with before v1.0:

• Pre-production: most AI use we hear about is in development and never reaches the screen, so the standard leaves it out. Is that the right line, or should some pre-production use be disclosed anyway? A model trained on a filmmaker's own work is one example that has come up.
• Assistive AI: would a platform, broadcaster or distributor actually use this category, and would any of them show it to viewers?
• Signing "No AI Used": software now switches AI features on by default, so should the declaration use a "reasonable enquiry" standard, the same one used for other chain-of-title warranties, where a producer attests to the best of their knowledge after checking?
• Animation and the in-between cases: in-betweening from an animator's keyframes and style transfer over hand-drawn frames. These build on human work but add something new, so they sit between our two questions. We want animators to weigh in.
• Co-productions outside the UK and US, where the paperwork is different.
• Marketing: trailers, posters and social cuts are out of scope for now. Tell us how much that matters.
• Display formats: which ways of showing the disclosure are practical and clear, from end credits to delivery metadata to an on-screen label. We will line the options up against the regulations as those firm up.

How to respond

Email contact@humanprovenance.film or use the GitHub repository, any time before 31 October. Partial answers are welcome: just say which question you are answering.