July consultation update

A month on from the first update. The definitions have not changed. The taxonomy page now has worked examples, a plain Markdown version, and a link for every section. What follows is what we heard, one request we are not acting on, and an answer on who controls this.

What we have heard

• Thresholds came up twice. A digital restoration artist and a festival submission platform both asked whether we should set a threshold for how much AI use triggers a category.
• The same artist worked through real restoration cases against the reconstruction test, in more detail than anyone so far.
• Festivals are writing their own AI policies, and they do not always map onto three categories. Submission platforms see this first-hand, because they carry whatever each festival decides.
• A submission platform said the real problem is filmmakers paying to enter festivals whose AI policy their film was never going to meet.
• The organising principle was tested on the distinction between tooling and expression.
• A submission platform and an organisation representing creative workers both doubted broad adoption, particularly at distribution and audience-facing layers.
• We are getting harder questions about neutrality, commercial value, and long-term governance.

Responses have now come from people and organisations in the US, Canada, the UK, and Europe.

On thresholds: heard, not yet acted on

Two people from very different parts of the industry asked for a threshold. We are recording it, but not acting on it for v1.0.

A threshold has to be measurable, and no tool in a normal pipeline currently reports what proportion of a frame, a shot, or a runtime an AI touched. We do not know when that changes. Writing the rule now would mean writing one nobody can apply.

So the current position stands. A production takes the highest category that applies to any element. We will revisit this when the tooling can support it.

On tooling and expression

An organisation representing creative workers challenged the organising principle on the difference between AI as a tool and AI as the source of the expression. If what you see on screen came from the model, the hours spent prompting may not change what it is, and copyright decisions appear to be heading the same way.

This supports the test. The taxonomy already says AI-generated content in the finished work is Generative AI whether or not a human could have produced it, and regardless of how the tool is marketed. Effort is not the test: what is in the finished film is.

Assistive AI describes what the AI did, not who authored the work, and v1.0 should leave no room to read it otherwise. None of this is legal advice, and productions should take their own.

On festivals and submission platforms

Festivals are drawing up their own AI policies, and they will not all map onto our three categories. Making the classification translate into a festival's own eligibility rules is our work, not theirs.

A submission platform told us it is unlikely to build verification that reads credentials on a finished film. The problem it sees is filmmakers paying entry fees for festivals whose AI policy their film was never going to meet. A classification can help with that. Work it out once, then check it against each festival's rules.

We have heard from submission platforms and need more from festivals: what your policy says, how you decide eligibility, and what would make our classification usable rather than another form to reconcile. If you programme or run submissions, get in touch.

On adoption, and the audience-facing layer

A submission platform and the organisation representing creative workers both doubted broad adoption, particularly at the distribution and audience-facing end. A disclosure standard that never reaches the surfaces audiences see is just paperwork moving between companies.

The EU AI Act's transparency obligations apply from 2 August 2026, covering deepfakes and AI-generated text on matters of public interest, with a lighter duty where the work is evidently artistic or fictional. That is a long way from disclosing how a whole film was made. The litigation we are watching is about training data and outputs, not disclosure.

So for now, we believe adoption has to happen voluntarily. That is why the standard is built the way it is: three categories, a producer self-declaration, free to use. Some of the people who doubt adoption told us the simplicity looks right.

We want it to end up where a viewer can see it: a platform's title information, a festival programme, end credits. In July, Netflix told shareholders that generative AI workflows were used in roughly 300 of its titles this year, with the largest concentration in post-production. Viewers of those titles have no indication of it. In June, A24 announced a research partnership with Google DeepMind, and the backlash from its own audience was immediate enough that the company defended the deal within days. Saying clearly what was and was not made with AI is more defensible than saying nothing and hoping the question does not come up.

On neutrality, commercial value, and governance

HPF was founded by The Mise En Scène Company, a film sales agency, which invites a reasonable question about whose interests it serves. What we have committed to:

• HPF describes what was used. It sets no rules about whether or how AI should be used in a production. Those judgements belong to festivals, guilds, distributors, and audiences.
• MSC holds the copyright and has licensed it under CC BY 4.0. That licence is irrevocable, so what has been published stays published. Anyone can use, adapt, or fork it.
• We are not building a verification or certification layer, and adopting the standard will not require payment. If other organisations want to build verification services on top of it, they should.

MSC is committed to finding a neutral home for the standard. The preferred outcome is an existing standards body adopting and maintaining it. Failing that, we would look to form a coalition or membership organisation so it is governed independently of any one company, including ours.

How to respond

If you work at a platform, broadcaster, or distributor, the questions we most want answered are what it would cost to carry the classification through ingest, and what would make displaying it worth doing.

The consultation runs to 31 October 2026. Email contact@humanprovenance.film or use the GitHub repository. Partial answers are welcome, just say which question you are answering.