Machine speed for checks. Human authority for decisions.
Accelerate submission triage, style-sheet checks, WCAG alt-text, and ONIX 3.1.2 hygiene with zero training on your manuscripts.
Clear boundaries between assistive checks and human locks
Every assistive feature is toggleable per title and gated behind human editorial, prepress, and finance sign-off.
| Workflow Domain | Assistive Diagnostic | Mandatory Human Gate |
|---|---|---|
| Submission Triage | Extracts chapter synopses, word counts, figure lists, and comparable cohorts | Commissioning editor owns acquisition decision |
| Editorial & Style | Flags timeline drift, Oxford/Chicago inconsistencies, and broken citations | Editor and author accept or reject each change |
| Prepress & Accessibility | Drafts WCAG 2.1 AA alt-text and flags TAC, bleed, and widow issues | Production lead signs off PDF/X-1a and EPUB 3.3 |
| ONIX 3.1.2 Metadata | Suggests Thema 1.5 qualifiers, BISAC codes, and retail copy variants | Metadata manager locks ONIX record before dispatch |
| Royalty Variance Checks | Flags unusual currency swings, missing feed periods, or reserve thresholds | Finance controller approves ledger and payouts |
Four technical controls protecting unpublished intellectual property
Stateless Ephemeral Inference
Inference payloads execute in memory inside your regional cloud boundary with zero prompt logging, zero caching, and zero weight updates.
Granular Contractual Opt-Out
When an author or literary agency contract prohibits automated text analysis, flipping one boolean flag on the Single Title Record disables AI endpoints for that work.
EU AI Act & Codelist Transparency
Every human editorial lock and accessibility alt-text review is logged with user ID and UTC timestamp for EU AI Act and EAA compliance audits.
Per-Tenant Envelope Keys
Manuscript files and proofs remain encrypted at rest under dedicated AWS KMS keys pinned to London, Frankfurt, Virginia, or Dubai.
Never. British Noble enforces zero data retention on inference endpoints and contractually prohibits training, fine-tuning, or indexing customer manuscripts, contracts, or sales data.
Our Responsible AI Charter enforces four pillars: human authorship protection, tenant and title opt-out switches, immutable provenance logs, and zero unauthorised ingestion.
No. Editorial diagnostics generate non-destructive side-by-side suggestions. Only an authorised human editor or author can accept changes.
Noble AI assists with submission synopsis extraction, comparable cohort sizing, Chicago/Oxford style consistency checks, citation verification, WCAG 2.1 AA alt-text drafting, and Thema 1.5 suggestions.
Acquisition approval, final editorial lock, contract execution, print release, and royalty disbursement always require authenticated human authorisation.
Yes. Administrators can disable AI across the entire tenant, per imprint, or on individual titles whose contracts restrict automated text analysis.
During EPUB 3.3 packaging, Noble AI drafts descriptive alt-text for figures and plates, placing every draft into an editorial approval queue.
Yes. Every Thema code, BISAC heading, or draft blurb records timestamped provenance showing which editor reviewed and locked the ONIX 3.1.2 field.
Inference routes strictly to regional endpoints matching your pinned data-residency zone (UK, EU, US, or UAE) over TLS 1.3.
Never. Our Acceptable Use Policy prohibits synthetic spam catalogues, protecting channel standing across all 108 distribution partners.
Deploy assistive publishing checks without compromising copyright.
Review our Responsible AI Policy or inspect per-ISBN governance controls live inside the 14-day sandbox.