Field notes / Provenance & Legal
AI disclosure and canon provenance for novelists and game developers.
Navigating platform disclosure requirements and intellectual property protection when using AI models requires distinguishing raw machine proposals from human-authored canon assets. Use this guide to review current platform policies and implement an auditable canon-state provenance workflow.
Understand platform disclosure rules and citations.
Major publishing and storefront platforms require creators to classify how AI models were used during production. Always verify current platform terms directly at submission time:
Steam / Valve Policy (January 2024): Valve requires game developers to disclose pre-generated AI content (art, code, text, audio) and live-generated AI content in the Steamworks AI survey, and to describe the guardrails used so live-generated AI content does not produce illegal content.
Amazon KDP AI Guidelines (September 2023): Amazon Kindle Direct Publishing distinguishes between 'AI-generated' content (text, images, or translations created by an AI-based tool—which KDP still classifies as AI-generated even if a human later applies substantial edits) and 'AI-assisted' content (content created by the author and refined, brainstormed, or edited using AI). KDP requires mandatory disclosure during title setup for all AI-generated content, while AI-assisted content does not require disclosure.
US Copyright Office Policy (March 2023): In its Copyright Registration Guidance for Works Containing Material Generated by Artificial Intelligence (88 FR 16190, March 16, 2023), the Copyright Office states that copyright protects only material that is the product of human creativity. Applicants must disclose appreciable AI-generated material so it can be disclaimed, while human-authored text and creative human selection, coordination, or arrangement remain eligible for registration.
Maintain a canon-state ledger for full auditability.
To prepare accurate platform disclosures and maintain clear editorial records, separate your production state into explicit tiers in a versioned continuity ledger:
- Raw Proposal (
canonState: 'proposal'): Model-generated text or image drafts. Logged with model version, prompt hash, and timestamp. Uncredited and non-binding.
- Human Curation & Transformation (
canonState: 'in_review'): Human author selects, edits, restructures, or rewrites the draft, adding character constraints and narrative direction.
- Accepted Canon (
canonState: 'accepted'): Binding asset recorded withauthorship: 'human_authored'or'human_transformed', human edit delta percentage, and platform disclosure category ('AI-generated'on KDP whenever the underlying draft was created by an AI tool, even after substantial human editing;'AI-assisted'only when the author wrote the work and used AI solely for editing or brainstorming).
Worked example: Canon-state workflow for a novel or game.
Consider developing a game dialogue sequence or novel chapter. In Step 1, an AI model proposes a 500-word character interaction. The workflow marks the entry as canonState: 'proposal' with model: 'claude-3-5-sonnet'.
In Step 2, the writer rewrites 350 words, inserts an original plot conflict, and alters the character's motivation, transitioning the scene through canonState: 'in_review' and recording the human transformation delta (70% human rewrite).
In Step 3, the asset transitions to canonState: 'accepted' and generates an export manifest. Because the underlying draft originated from an AI tool, the manifest still flags the scene for mandatory 'AI-generated' disclosure on Amazon KDP (while documenting the 70% human rewrite for editorial provenance), whereas a scene drafted by the human author and only proofread with AI is recorded as 'AI-assisted'. On Steam's AI survey, any retained pre-generated AI assets are logged under pre-generated AI disclosure.
Export an auditable provenance packet with UTM tracking.
When preparing your book or game for release, export plain Markdown and JSON metadata documenting your canon-state history. These supporting production records help teams document human editorial revisions and prepare consistent platform disclosures, though they do not constitute legal advice or guarantee copyright eligibility.
Test your project using our open tools and production kits with captured tracking links: try the Free World Starter or inspect the full Worldbuilder Production Edition.
Try the worked example.
Read the free demonstration scene and download the editable World Starter. The example is an AI-assisted proposal outside official Arcanea canon; adapt its method to your own original project.
Open the free World StarterInspect the Production Edition