Auto-classification by content
Markings, personal data, research vocabulary and aggregation rules (unit + location + date) propose a level with evidence highlighted in the text. Humans confirm RESTRICTED, R&D and low-confidence cases.
An interactive blueprint of an enterprise knowledge and document lifecycle platform for international organisations. Content-based classification by network level, attribute-based access control, AI first review with human approval, and guardrails that keep AI agents out of administration.
Open the interactive demoWhat it showsMarkings, personal data, research vocabulary and aggregation rules (unit + location + date) propose a level with evidence highlighted in the text. Humans confirm RESTRICTED, R&D and low-confidence cases.
Each level lives on its own network domain — Internal, RESTRICTED and an R&D enclave — with its own index, models and keys. Anything above RESTRICTED is blocked at upload and handed to a separate Mission Critical System.
Clearance, need-to-know, department, network and device decide every search, view, share and download live. Hidden content is never shown, counted or passed to AI.
Scorecard for tone, structure, template coverage and risk; side-by-side diff; 80% minimum and 85% fast-track thresholds; multi-level approval chain with justified overrides.
Agents are non-human identities with tool manifests. They can never create, delete or modify users, assign roles, set classification, approve or call external URLs. Prompt injection is detected and neutralised.
Every published version is sealed with a SHA-256 checksum; changes outside the workflow lock the document. The audit log is append-only and hash-chained.
Review dates notify owners. Lowering a level needs a request, an AI risk assessment including aggregation, and a signed Security Officer approval through a cross-domain guard.
Levels per network, access decisions, alerts, integrity status and a risk register derived from incidents and behaviour.
Draft → review → published → archived → purged. Hot, Warm and Cold tiers per network, retention schedule, legal hold and crypto-shredding for erasure.
Every published version is kept with its own level and checksum; compare any two versions word by word and roll back through an audited step.
Entities and relations per network; path questions are checked for aggregation and held for the Security Officer when they combine sensitive elements.
Department profiles, a live EN/FR glossary and an acronym register drive the First Reviewer’s suggestions.
Quizzes generated from approved policies, inheriting the source level; institutional voice and profiles in one place.
Approvals are signed (simulated QES), each step has a deadline, and out-of-office delegation or escalation keeps work moving.
Device, session risk and working-hours rules with dual-control changes and a policy tester.
Simulated IdP, M365, Google, email, ERP and legal connectors through the inbound gateway; CEF export to SIEM; offline vault and hash-verified delta packages.
Executive and storage dashboards, non-functional targets, and a compliance matrix mapped to ISO 27001, NIST, EU AI Act, GDPR and eIDAS.
Department and project spaces; sharing to people, groups or departments with a live policy check that shows who is excluded and why.