Data protection
A commercial overview of how Node16.ai approaches institutional data control, human oversight and information security.
Raw transaction records, customer identities and bank-local investigation evidence remain in the bank-scoped environment. Shared learning is designed around governed model contributions, rather than a central pool of bank transaction data.
Banks derive pseudonymous linkage tokens from verified customer or entity identifiers using keyed cryptographic functions. Matching tokens can link records belonging to the same underlying entity without sharing names or original identity numbers in the linkage exchange. Identity information and cryptographic key custody remain subject to bank-governed controls. Link quality depends on the banks’ verified identity data and common normalisation rules; access to further case detail follows the disclosure workflow.
The bank-to-bank secure-aggregation design uses ML-KEM-768, standardised in NIST FIPS 203, to establish the shared key material used for cryptographic masking. It is designed to address the risk of an adversary recording these exchanges now and attacking them with a future quantum computer.
Sonar combines GNN-based detection and bank-local pattern analysis with differential privacy as the production model-release control. This limits the information a released model should reveal about a person’s contribution. The privacy statement and its evidence are qualified for the particular mechanism and release.
Node16.ai coordinates the service while each bank retains responsibility for its participation and authorised use. The partnership arrangements define participant roles, approval of model changes, access to audit evidence and supervisory visibility. Model versions and attributable decisions support that institutional oversight.
Learning from a shared model is different from sharing information about a particular case. Model release, pseudonymous linkage and collaborative disclosure each have their own access, purpose and release controls.
Investigator brings evidence, draft narratives and workflow support to the analyst. Review, escalation and reporting decisions remain attributable to the authorised people. Risk tolerance, roles and the audit trail are part of the product design.
Company mail, CRM and internal AI tools handle business relationships and company work. Bank customer records, transaction data and case material do not belong in those tools. Product access and corporate-tool access are kept separate.
Our ISO/IEC 27001 programme is in progress. It covers risk management, access, suppliers, change control, continuity and retained evidence. Our own programme and the assurance of our infrastructure providers are assessed separately.
Data placement, access, retention, recovery, suppliers and contractual responsibilities are addressed for the intended engagement. Model training and evaluation use sovereign supercomputing infrastructure, with product data kept in bank-scoped environments; the deployment and service arrangements are part of bank diligence.
The Danish FSA's AMLR Article 75 sandbox is where we clarify the relationship between AMLR, GDPR and the AI Act. We are aiming for commercial clarity in Q1 2027 and developing Collaborator under Article 75 for July 2027.
Discuss the approach with us.
Request a briefing →