Ordellum PlatformEcosystem
Trust and human oversight
Every Ordellum product produces decision support. Irreversible actions always remain with a person.
However advanced the AI models, rule engines and cryptographic systems become, no single component is allowed to act alone. This principle is built into every decision point as a systematic safeguard against the risk of excessive automated authority, in line with the OWASP guidance on “excessive agency” for AI systems.
In Analytics, a final decision on a high-risk case requires at least a two-thirds agreement across independent AI models together with a human analyst. In the AI Native SDK, no mainnet deployment can start without explicit user approval, a security check and a deployment summary.
Where responsibility sits
- Screens, classifies and scores transactions and counterparties
- Stays with you: Deciding whether to block, release or report
- Prepares evidence bundles and draft suspicious-activity reports
- Stays with you: Approving and filing reports with the authorities
- Prepares Travel Rule data and audit records
- Stays with you: Running VASP-to-VASP messaging and transfer decisions
- Generates, verifies and packages code with signed evidence
- Stays with you: Approving every mainnet deployment and its gas costs
Ordellum Analytics is a software decision-support layer. It is not a VASP or CASP, does not execute transfers and does not file reports with authorities on a client’s behalf.
Security is treated as a quality requirement embedded in every stage, not as a final inspection step. In the AI Native SDK, the security perspective holds a veto that can override other agents, so nothing reaches production below the required security standard.
Automated security reports are an important safeguard, but they are presented for what they are: they do not replace independent human review or a formal smart-contract audit.
- Prove, don’t disclose
- Zero-knowledge proofs let people prove facts such as age or residency without revealing the underlying personal data.
- Minimal retention
- Raw identity documents are not retained by default; only cryptographic commitments and proof metadata are kept.
- Masked before AI
- Personally identifiable information is tokenised before any context reaches an AI model, and customer data is never used to train models.
Products move to production through staged release gates rather than in a single step. A later stage cannot begin until the current one is proven with evidence.
P0
Documentation and threat model
Architecture decisions, threat modelling, data-protection impact assessment and risk registers are completed and signed off.
P1
Closed pilot
A limited group of design partners use the product in shadow mode so that false positives and negatives can be measured.
P2
Controlled general availability
Load, chaos and penetration testing, signed builds and rehearsed runbooks open the standard plans to customers.
P3
Scaled general availability
Thirty days of service-level history, cost validation and disaster-recovery drills unlock high-volume enterprise contracts.