Tobie Morgan Hitchcock, Co-founder and Chief Executive Officer, SurrealDB
Financial institutions are deploying AI faster than the data infrastructure beneath it can support. This paper makes the architectural case that the dominant failure mode of financial AI is not the model - it is the fragmented data estate underneath it. It sets out the three architectural commitments a trustworthy financial AI platform requires, and how a unified, transactional, governed foundation satisfies SR 11-7, the EU AI Act, BCBS 239, MiFID II, and DORA structurally rather than procedurally.
Key takeaways
Why only 2 of 31 global systemically important banks are fully BCBS 239 compliant after twelve years - and why the root cause is architectural, not procedural.
The three non-negotiable commitments a financial AI platform requires: a unified multi-model substrate, structured agent memory with provenance, and governance by design.
How bi-temporal data, record-level access control, and write-back audit trails satisfy SR 11-7, the EU AI Act, BCBS 239, MiFID II, and DORA structurally rather than procedurally.
Five questions every institution should answer before choosing between open infrastructure and closed AI platforms.
A reference architecture mapping each layer - from object storage to reasoning model - to the financial AI capability it enables.
What's inside
A note from the CEO
Why financial AI keeps failing at the foundation
Why financial data is structurally different
What the architecture of trust requires
Spectron: agent memory built into the database
SurrealDB v3.1: the current platform
Open infrastructure vs. closed platforms: five questions
The reference architecture
Four principles for the AI age in financial services
Key regulatory frameworks referenced
Read the whitepaper
Enter your email to read “The architecture of trust”.
GET STARTED
Build on the context layer
Documents, graphs, vectors, and time-series in one ACID transaction - from object storage to agent memory.
SOC 2 Type 2
GDPR
Cyber Essentials Plus
ISO 27001