DRAFT, pending counsel review. This document is an internal draft prepared on 2026-04-26 by the engineering team. It has NOT been reviewed by external legal counsel. Do not rely on it for legal advice. Effective date is a placeholder pending sign-off. Apostle Pty Ltd makes no representation that this draft satisfies any specific jurisdictional requirement until counsel-reviewed.

AI Disclosure Policy

Last revised: 2026-04-26 · Version: 0.1.0-draft

Table of contents

1. Why this policy exists
2. Scope
3. Definitions: what counts as "AI use"
4. Mandatory disclosure at submission
5. The six-criterion human-creative-contribution rubric
6. Provenance fields captured (ai_provenance table)
7. Prohibited categories
8. Display obligations on the title detail page
9. Penalties: removal, strikes, and suspension
10. Copyright enforceability disclosure
11. Audit, evidence and retention
12. Future evolution and the AI tools list
13. Cross-references
14. Contact
15. Version history

1. Why this policy exists

PYLON is a curated platform for short-form narrative AI-assisted cinema. We accept films where humans direct generative tools, and we decline films where the human role is reduced to a prompt. The point of this policy is to make that line legible to filmmakers, predictable for curators, and defensible to platforms, regulators and rights-holders.

Two regulatory frames inform the document:

• The United States Copyright Office's Copyright and Artificial Intelligence, Part 2: Copyrightability (January 2025). The Office reaffirmed that copyright protection turns on human authorship and that purely machine-generated material is not copyrightable.
• The Australian Online Safety Act 2021 and the eSafety Commissioner's expectations of platforms hosting synthetic media, including the basic online safety expectations and class-1A / class-1B classification.

Where a work lacks sufficient human creative contribution to be copyrightable in the United States, this affects what we can offer in our licence, see §10 below.

2. Scope

This policy applies to:

• Every title submitted via /creator/submit.
• Every asset uploaded to PYLON for inclusion in a title (images, audio, captions, posters, stills).
• Every reupload, re-cut or supplementary asset for a published title.

It does not apply to user-generated commentary in Q&A threads or watch parties, that is governed by the Community Guidelines at /legal/community-guidelines.

3. Definitions: what counts as "AI use"

For the purposes of this policy, "AI use" means any of the following:

• Generation. Producing image, video, audio or text frames using a model that synthesises output from a prompt or seed (text-to-image, text-to-video, image-to-video, text-to-audio, voice-clone, music generation, lip-sync, motion-capture transfer, neural compositing, upscaling beyond restoration, frame interpolation beyond clean cadence, generative outpainting, generative inpainting beyond spot-cleanup).
• Voice or face cloning of a real person (including the filmmaker themselves) using a model trained on that person's data.
• Style transfer that materially changes the source, for example, passing a live-action plate through a model that re-paints every frame in a different visual idiom.
• Output of a model trained on copyrighted material without licence where the output substantially reproduces protectable elements of the training corpus.
• Tool-chain compositing, generative B-roll cut against live-action primary footage is in scope and must be disclosed.

Routine post-production tools that do not synthesise frames, colour grading, audio EQ, conventional VFX, mask-based rotoscoping, optical flow, tracking, grain matching, are not in scope. If you are unsure, disclose; we'd rather over-collect provenance than miss it.

4. Mandatory disclosure at submission

Every Title submitted to PYLON must include an AI provenance attestation. The attestation is collected at submission time via /creator/submit/provenance and is persisted to the ai_provenance table (packages/db/src/schema/content.ts). Submission cannot proceed without it.

The attestation is a sworn statement by the submitting creator, signed under penalty of perjury (US) or by statutory declaration (AU equivalent under the Statutory Declarations Act 1959 (Cth)). The full text of the sworn-statement form is at /legal/ai-provenance-attestation.

False statements made in this attestation (a) void the licence the creator grants PYLON under the Filmmaker Distribution Agreement at /legal/filmmaker-distribution-agreement, (b) trigger immediate title removal, and (c) may expose the creator to civil liability and to criminal liability under §11 of the Statutory Declarations Act 1959 (Cth) (up to four years' imprisonment).

5. The six-criterion human-creative-contribution rubric

PYLON uses a six-criterion rubric to determine whether a film demonstrates "sufficient human creative contribution" to be eligible for publication. Each criterion is scored 1–5 by the assigned curator. The rubric is published in full at /editorial-policy. The six criteria are:

1. Visible selection. Evidence that the human creator chose among generated material, kept some takes, discarded others, made judgments about which generated frames serve the story.
2. Arrangement. Sequencing decisions: cut order, pace, in/out points, juxtaposition. Arrangement is itself a copyrightable contribution under US compilation-authorship doctrine.
3. Iteration. Multiple drafts demonstrably reviewed and revised. We treat iteration evidence (project files, version history, timestamped renders) as a strong signal of authorship.
4. Direction. Specific creative intent documented in the submission's human_contribution_statement. "I gave a prompt" does not satisfy direction. "I directed for X tone, rejected Y takes for Z reasons" does.
5. Post-production work. Conventional craft, colour, sound design, final mix, conform, grade, performed by the human creator on top of generated material.
6. Final cut authority. A human creator made the last decision about what is in the film and what is not. A film that was assembled by an automated pipeline without a final human pass does not satisfy this criterion regardless of its other strengths.

Films must score a combined 22 of 30 to publish. Films at 18–21 escalate to a senior curator. Films below 18 are declined. The six-criterion structure intentionally separates the fact of human creative contribution (relevant for copyrightability) from the quality of the film (relevant for editorial selection).

6. Provenance fields captured (ai_provenance table)

At submission we persist the following fields in the ai_provenance table:

• tools_json, JSON array of AI tools used, by name and version, with the role each played (image generation, video generation, upscaling, voice clone, etc.).
• human_retouch_pct, integer 0–100. The creator's good-faith estimate of the share of frames touched by hand after generation.
• human_contribution_statement, required free-text statement describing the human creator's role: selection, arrangement, iteration, direction, post, final cut.
• source_assets_json, JSON array of source-asset licences (training data licences where applicable, original assets, public-domain or Creative-Commons items with attribution).
• real_person_likenesses, boolean. True if any identifiable real person appears.
• real_person_consent_note, free-text reference to consent documentation (release id, talent agreement) when real_person_likenesses is true.
• flagged / flag_reason, internal review flags.

These fields are retained for the longer of (a) the lifetime of the title on PYLON and (b) seven years from withdrawal, in support of DMCA defence, copyright disputes, and platform safety reviews.

7. Prohibited categories

PYLON will not publish, and will remove if discovered post-publication, any work that includes:

• Non-consensual deepfakes of any identifiable person, living or recently deceased, without documented written consent on file. This includes any synthetic depiction of a real person in a sexual, defamatory, or otherwise reputation-damaging context regardless of consent.
• Child sexual abuse material (CSAM) of any kind, including synthetic depictions of minors in sexual contexts. We treat synthetic CSAM identically to non-synthetic CSAM and report to relevant authorities (NCMEC in the US; the AU eSafety Commissioner under the Online Safety Act 2021).
• Synthetic voice or face impersonation of public figures, heads of state, candidates for public office, judicial officers, in a manner that could mislead a reasonable viewer about the figure's actual conduct or speech.
• Election misinformation generative content, synthetic media that depicts a candidate, a poll, an electoral official, or an electoral process in a manner intended to deceive voters about the time, place, manner, eligibility, or integrity of an election.
• Identity- or hate-targeted synthesis, synthetic content created to harass, dehumanise, or incite violence against an individual or protected class.
• Trademarked-character infringement, synthetic content reproducing protectable elements of trademarked characters or worlds without licence.

This list is non-exhaustive. PYLON reserves the right to decline any title that, in its editorial judgment, conflicts with the values expressed in this policy.

8. Display obligations on the title detail page

For every published title, the title detail page surfaces a concise AI provenance summary derived from titles.ai_provenance. The summary includes:

• a list of AI tools used (by name);
• the human-retouch percentage;
• a one-sentence excerpt of the human_contribution_statement;
• a flag for synthetic real-person likenesses (with the consent note reference, where applicable);
• the curator's published rubric score-band (<18 / 18–21 / 22–30).

This UI surfaces only what a viewer needs to make an informed choice; the full provenance record is retained internally for legal and editorial use, not for public display.

9. Penalties: removal, strikes, and suspension

A creator who submits a title that violates this policy faces the following operational consequences:

• Removal. The title is unpublished and the playback signing key is revoked.
• Strike. The creator's dmca_strike_count (used as our general-purpose moderation strike counter; see schema comment) is incremented for any prohibited-category violation that also implicates copyright or right of publicity.
• Editorial cooldown. Three declined submissions in a 12-month window trigger a six-month resubmit cooldown (see Editorial Policy at /legal/editorial-policy).
• Suspension. Three strikes within twelve months result in account suspension; see Trust & Safety Policy at /legal/trust-and-safety-policy for the appeals process.

Material in the prohibited categories of §7 is treated as immediately removable on first instance. We do not apply a "warn first" model to CSAM, non-consensual sexual deepfakes, or election-misinformation synthesis.

10. Copyright enforceability disclosure

The United States Copyright Office's January 2025 Part 2 report confirms that a work lacking sufficient human creative contribution is not copyrightable in the United States. PYLON cannot extend a US copyright that the law does not recognise.

For titles where the rubric score does not establish sufficient human authorship, our licence in the Filmmaker Distribution Agreement is limited to the PYLON surface itself. We cannot, and do not, guarantee exclusivity against third-party redistribution of non-copyrightable elements; we can guarantee that we will not knowingly publish a duplicate.

In Australia, originality and substantial form are the relevant tests under the Copyright Act 1968 (Cth). The Australian position on AI authorship is unsettled and the test for whether a work has been "made" by a "qualified person" remains under examination by the courts; we follow the conservative reading.

11. Audit, evidence and retention

Every submission, every rubric score, every curator decision, every withdrawal, and every appeal outcome is recorded in the immutable audit_log table with actor, timestamp and reasoning. The ai_provenance row for a published title is retained for the longer of the title's lifetime on PYLON and seven years from the date of withdrawal.

We rely on this evidence to:

• defend §512 takedowns and counter-notices;
• support copyright-enforceability statements made to licensees;
• substantiate our editorial decisions in any later dispute;
• comply with the AU eSafety Commissioner's reporting obligations under the Online Safety Act 2021.

12. Future evolution and the AI tools list

The list of AI tools considered in scope for disclosure is maintained at /legal/ai-tools-list (TBD). It is amended at least annually and on a rolling basis as new generative tools enter common filmmaker workflows.

This policy itself is reviewed annually. Material changes are versioned in the version history below; non-material changes (typo fixes, link updates) are tracked in the underlying repository commit history but do not bump the policy version.

13. Cross-references

• Filmmaker Distribution Agreement at /legal/filmmaker-distribution-agreement
• AI Provenance Attestation (sworn-statement template) at /legal/ai-provenance-attestation
• Editorial Policy (rubric in full) at /legal/editorial-policy
• Trust & Safety Policy (appeals) at /legal/trust-and-safety-policy
• DMCA Policy at /legal/dmca

Contact

• AI provenance questions: [email protected]
• Editorial: [email protected]
• Legal: [email protected]

Version history