One question. Two architectures. Two answers.
The three eras of identity verification
The Money Laundering Regulations did not change. The operational reality of complying with them did. Below, the PII custody surface across three eras - visualised as the density of obliged entities holding sensitive identity data, and the failure mode each era introduced.
Verification at the bank counter
Identity work concentrated at institutional choke points. A few entities equipped to do the work: compliance teams, secure infrastructure, statutory accountability.
Not architectural risk.
Document collection moves to every node
The pandemic moved the economy remote. Every estate agent, every accountant, every small lender now collecting and holding sensitive identity documents on consumer-grade infrastructure.
The £38.3bn compliance bill begins.
Synthetic documents bypass the checkpoint
Generative AI produces fully synthetic passports, statements, payslips. The custody surface that arose in Era II is now the surface AI defeats most reliably - the entities receiving the documents cannot detect the forgeries.
8.3% of digital onboarding flagged suspicious, H1 2025.
Where does the trail go?
AML's purpose is to enable the investigation and tracing of illicit value. When fraud is detected later, the regime's value depends on having somewhere to trace it back to. Two architectures, the same question, different answers.
Document-centric path
The investigator follows the trail upstream. It reaches the document accepted at the moment of onboarding. The document is the only substance. If it is synthetic, the trail ends there.
Witnessing path
The investigator follows the trail upstream. It passes through a Broker Certificate, into the customer's authenticated relationship with a regulated institution, then into the institution's own statutory records, ongoing monitoring, and supervised oversight.
What the witness model looks like
The platform observes - passively, without retrieving documents, without contacting institutions - that the customer demonstrably controls an authenticated relationship with a regulated institution. Four independent signals converge on a single output.
The custody surface, before and after
The same set of obliged entities; the same regulatory framework; two different architectures for where the sensitive identity data physically lives. The honeypot density is not a metaphor - it is the literal count of nodes holding PII.
Every obliged entity is a custody node
One custody node. The rest hold nothing.
The framework already supports witnessing-not-custody. The architecture exists. The adoption has not yet caught up.
The diagrams above are not a future state. They describe an operating platform working within the UK Money Laundering Regulations 2017 and the Data (Use and Access) Act 2025 - as those instruments are written today.