Binding BISG-Aligned Responsible AI & Copyright Charter
Contractually enforceable technical, cryptographic, and editorial guardrails protecting human authorship, publisher copyright, and literary agency riders.
Executive Summary & Constitutional Copyright Pledge: Human imagination, scholarly inquiry, and authentic authorial voice are the irreplaceable foundation of global book publishing. Unlike consumer generative portals that treat uploaded prose, cover artwork, and studio voice recordings as uncompensated training corpora, British Noble enforces a legally binding, cryptographically verified Responsible AI & Copyright Charter aligned with Book Industry Study Group (BISG) AI best practices, UK Publishers Association principles, Society of Authors and Authors Guild transparency standards, and the European Union Artificial Intelligence Act (Regulation (EU) 2024/1689).
Incorporated directly as an enforceable warranty under our Master Terms of Service, this Charter establishes five non-negotiable guarantees across every British Noble AI and British Noble Press workspace: (1) Absolute Zero Model Training on customer manuscripts, proofs, artwork, audio stems, ONIX metadata, or financial ledgers; (2) Stateless Zero-Data-Retention Inference pinned strictly inside Customer's sovereign KMS cloud vault; (3) Three-Tier Cryptographic Kill-Switches enabling instant opt-out at tenant, imprint, or individual Title/ISBN level to honour literary-agency riders; (4) Mandatory Human Sign-Off Gates preventing autonomous prose replacement or autonomous publishing; and (5) Immutable Per-Title Provenance Certificates supporting agency, university, and retail channel audits.
1. Constitutional Primacy of Human Authorship & Binding Status
This Responsible AI & Copyright Charter (BN-GOV-RAI-2026) is not an aspirational marketing manifesto; it constitutes an enforceable contractual warranty incorporated directly into Clause 5 of our Master SaaS & Author Studio Terms of Service and Annex II of our Enterprise Data Processing Addendum. Any breach of the zero-training or confidentiality covenants herein triggers the Enhanced Tier 2 Super-Cap remedies under the Agreement.
British Noble's architecture draws a bright operational line between mechanical publishing diagnostics (where deterministic rules and optional assistive checks save editors and production controllers hundreds of repetitive hours checking Oxford/Chicago style drift, broken cross-references, missing WCAG alt-text, or Thema subject codes) and human creative authorship. Creative expression, scholarly thesis construction, commissioning judgement, cover artistry, and audiobook performance belong exclusively to human creators and their publishing houses.
2. Absolute Prohibition on Foundation Model Training & Fine-Tuning
British Noble covenants irrevocably that neither British Noble nor any infrastructure or inference sub-processor shall ever use, copy, sample, scrape, vector-index across tenants, license, or retain any Customer Content to train, pre-train, fine-tune, reinforce (RLHF/RLAIF), distil, calibrate, or evaluate any machine-learning model, large language model, diffusion image model, or synthetic speech engine.
This absolute prohibition covers every byte uploaded to or generated inside a workspace, including: (a) unsolicited submissions, book proposals, and reader reports; (b) draft chapters, developmental notes, copyediting redlines, and page proofs; (c) bespoke cover artwork, photography, architectural drawings, and interior typography; (d) studio audiobook recordings and narrator voice stems; and (e) acquisition P&L models, author contracts, retail discount schedules, and royalty ledgers. One publisher's manuscripts or editorial decisions can never influence, leak into, or inform suggestions surfaced inside another publisher's tenant.
3. Stateless Zero-Data-Retention (ZDR) Regional Inference
Whenever an authorised user invokes an optional assistive workflow inside Noble AI (such as scanning a thirty-chapter academic volume for inconsistent hyphenation or drafting initial WCAG 2.1 AA alt-text descriptions for complex charts), processing executes exclusively on enterprise-isolated, Zero-Data-Retention (ZDR) compute endpoints deployed inside Customer's selected sovereign cloud region (UK eu-west-2, EU eu-central-1, US us-east-1, or UAE me-central-1).
Under ZDR architectural enforcement: (a) input text or figure crops exist solely in encrypted volatile RAM for the duration of the synchronous API call and are zeroed immediately upon response transmission; (b) zero prompt logs, completion caches, or abuse-monitoring snapshots are persisted to disk by underlying compute providers; and (c) zero third-party vendor employees or contractors are permitted to inspect prompts or outputs under any circumstances.
4. Cryptographic Tenant, Imprint & Per-Title AI Kill-Switches
Leading literary agencies and author estates increasingly include explicit contractual clauses prohibiting any automated AI processing of a contracted work without the author's prior written consent. Spreadsheets cannot enforce those clauses reliably across editorial and production desks; British Noble enforces them directly in software via a three-tier hierarchical control plane:
- Level 1 - Tenant-Wide Global Disablement: CISO or Managing Director administrators can disable all neural assistive endpoints across the entire organisation with a single toggle while keeping 100% of deterministic relational workflows, 42-rule preflight checks, ONIX 3.1.2 validation, and royalty calculations active.
- Level 2 - Imprint & Workflow Policy Locks: Publishing groups can permit assistive metadata or WCAG alt-text drafting on a digital reference imprint while locking prose-facing assistive tools completely off across a literary fiction or poetry imprint.
- Level 3 - Per-Title / Per-ISBN Contractual Kill-Switch: Checking
Restrict Assistive AI (Agency Rider)on a Single Title Record's contract screen cryptographically blocks API routing of that Title's manuscript, images, and audio to any neural model endpoint across every user role in the workspace.
5. Permitted Assistive Diagnostics vs Strictly Prohibited Uses
To maintain catalogue quality across 108 global retail and library channels and protect human creators, British Noble establishes clear boundaries between supported assistive diagnostics and strictly prohibited platform abuses:
| Governance Category | Permitted Assistive Workflows (Human-Reviewed) | Strictly Prohibited Actions (Blocked by Platform & Contract) |
|---|---|---|
| Editorial & Manuscript | Flagging Chicago 18th / Oxford spelling drift, timeline inconsistencies, broken citations, and structural pacing diagnostics as non-destructive margin notes | Mass-generating synthetic "spam" books, plagiarising third-party works, or silently rewriting authorial prose without tracked human review |
| Accessibility & Metadata | Drafting candidate WCAG 2.1 AA image alt-text, suggesting EDItEUR Thema 1.5 / BISAC codes, and checking ONIX 3.1.2 blurb completeness | Publishing unverified synthetic accessibility claims or deceptive keyword-stuffed retail metadata |
| Cover Art & Photography | Deterministic DPI upscaling warnings, CMYK total-area-coverage (TAC) ink diagnostics, and spine calliper geometry checks | Ingesting illustrator portfolios to mimic living artists' styles or stripping C2PA / IPTC creator attribution metadata |
| Audio & Voice | Automated LUFS loudness metering, room-tone floor checks, breath/click flagging, and text-to-audio proof alignment | Cloning or synthesising any human narrator's or author's voice without an explicit, verified, standalone biometric voice licence |
| Acquisitions & Finance | Comparable-ISBN market sizing tables and statistical anomaly alerts on messy retailer CSV sales imports | Autonomous algorithmic rejection of author submissions without human editorial review, or autonomous bank disbursements |
6. Non-Destructive Redlines & Mandatory Human Sign-Off Gates
Every assistive diagnostic inside British Noble is engineered on a non-destructive, human-in-the-loop architecture. Assistive checks surface solely as attributed margin comments, tracked suggestions, or candidate metadata chips alongside confidence context and source line references; no assistive feature possesses write-lock authority over canonical publication fields.
Authenticated human sign-off by an authorised workspace user is technically mandatory before executing six critical lifecycle transitions: (1) recording a formal commissioning offer or rejection; (2) locking an edited manuscript version for typesetting; (3) marking WCAG 2.1 AA image descriptions as editorially verified; (4) approving print-ready PDF/X-1a plates, cover spreads, or EPUB 3.3 masters; (5) transmitting ONIX 3.1.2 feeds to retail partners; and (6) closing a royalty accounting period or releasing payee payouts.
7. Immutable Per-Title AI Provenance & Audit Logging
Transparency requires verifiable evidence rather than verbal assurances. Every single invocation of an assistive feature inside a workspace writes an immutable cryptographic entry to the Title's AI Provenance Ledger, recording: UTC timestamp, authenticated user ID, Single Title Record ISBN, specific assistive function invoked (e.g., OxfordStyleConsistencyCheck or AltTextCandidateDraft), pinned cloud region, model identifier and version, and the human editor's explicit disposition (Accepted, EditedBeforeAccept, or Dismissed).
Conversely, where a Title has its Per-Title AI Kill-Switch enabled from ingestion through publication, workspace administrators can generate a signed Zero-AI Provenance Certificate proving mathematically from immutable WORM audit logs that zero neural inference calls ever touched that ISBN during its lifecycle, satisfying literary-agency, prize-jury, and scholarly press audits effortlessly.
8. EDItEUR ONIX 3.1.2 & Retailer AI Disclosure Alignment
Global retail channels (including Amazon KDP, Ingram, Apple Books, and European library aggregators) as well as EDItEUR ONIX 3.1.2 codelists distinguish sharply between AI-assisted diagnostic workflows (using software for spellchecking, Oxford/Chicago style verification, citation linting, accessibility tagging, or structural diagnostics on a human-written work, which does not constitute synthetic authorship and does not trigger consumer AI-generated labels) and AI-generated content (text, images, translations, or audio created autonomously by generative prompts).
Where a digital or hybrid publisher lawfully commissions a declared synthetic asset (such as an explicitly licensed synthetic audio edition for out-of-print technical reference manuals where all contributor rights are cleared), British Noble's Distribution Network provides native ONIX 3.1.2 provenance elements and retailer intake flags so downstream bookstores and library catalogues receive accurate, standards-compliant transparency metadata without miscategorising human-authored books. Conversely, British Noble enforces automated TDM (Text and Data Mining) reservation tags inside compiled EPUB 3.3 package metadata (including W3C TDM Reservation Protocol tdm:reservation="1" headers and ONIX Codelist 145 usage constraints) so published digital editions explicitly signal statutory opt-out from unauthorised third-party web-crawler scraping under Article 4(3) of the EU Digital Single Market Copyright Directive (Directive (EU) 2019/790).
9. Narrator Biometric Voice & Illustrator Moral Rights Protections
British Noble upholds equity-aligned contractual and technical protections for human audiobook narrators, literary translators, jacket designers, documentary photographers, and fine-art illustrators. Studio audiobook masters uploaded to Book Production Studio are processed exclusively inside ephemeral acoustic preflight containers for deterministic signal validation (ITU-R BS.1770-4 LUFS loudness, -3.0 dBTP true-peak ceiling, 44.1 kHz / 24-bit PCM integrity, noise-floor consistency, and chapter header timing) prior to direct retail delivery across Audible, Spotify, and Apple Books; narrator voice stems are never exposed to voice-cloning or speech-synthesis models.
Similarly, uploaded cover designs, museum plates, architectural blueprints, and 254 x 305 millimetre coffee-table photography retain intact embedded ICC colour profiles, Coalition for Content Provenance and Authenticity (C2PA) cryptographic credentials, and IPTC creator copyright metadata wherever supported by print and digital output specifications, and are never indexed into visual generation datasets or cross-tenant asset libraries.
10. Compliance with EU AI Act, UK CDPA & US Copyright Standards
British Noble's AI Governance Committee (comprising General Counsel, the CISO, the DPO, and Principal Publishing Systems Architects) audits platform capabilities quarterly against evolving statutory, regulatory, and judicial frameworks across all four operating regions, including: Regulation (EU) 2024/1689 (the EU Artificial Intelligence Act, confirming zero prohibited practices under Article 5, clear low-risk B2B assistive classification outside Annex III high-risk profiling, and complete deployer transparency technical files), the UK Copyright, Designs and Patents Act 1988 (CDPA), US Copyright Office human-authorship registration guidance (37 CFR Part 202), and DIFC intellectual-property regulations.
Before any external enterprise inference endpoint is approved as an optional sub-processor inside our sovereign vaults, Legal Counsel and the CISO verify legally binding zero-retention enterprise addenda, absence of third-party copyright infringement claims impacting inference outputs, and regional compute pinning.
11. Customer Verification Rights, Agency Riders & AI Governance Desk
Enterprise customers, literary agencies conducting technical due diligence alongside client publishing houses, and university press editorial boards may inspect our Zero-Data-Retention sub-processor architecture attestations, W3C TDM opt-out EPUB samples, and sample Per-Title Zero-AI Provenance Certificates under mutual confidentiality agreement by contacting Legal@BritishNoble.com.
To assist publishing legal departments updating their standard author agreements, our AI Governance & Legal Counsel Desk also provides complimentary model contract schedule language aligning publisher-author AI undertakings directly with British Noble's Three-Tier Kill-Switch controls. Any future material addition to assistive capabilities is documented thirty (30) days in advance in enterprise release notes and defaults strictly to disabled (opt-in only) across existing enterprise workspaces.
Multi-Jurisdiction Governance
Aligned with UK GDPR, Data Protection Act 2018, EU SCCs, CCPA/CPRA, UAE Federal Decree-Law No. 45 of 2021 (PDPL), ISO/IEC 27001:2022, and SOC 2 Type II.
Zero Manuscript Training
Customer manuscripts, proof files, ONIX records, and royalty ledgers are isolated per tenant and never used to train foundation or shared models.
AI Governance & Legal Counsel Desk
Enterprise procurement teams may request countersigned schedules, DPA exhibits, and audit certificates directly.
Legal@BritishNoble.com →