AI & Compliance Systems
AI-enabled KYC assurance that keeps the decision with the human
Cognitive Sentinel is our approach to applying explainable AI to KYC assurance — identifying weak, missing or unsupported evidence, explaining the basis for challenge, and leaving every final decision with an accountable human reviewer.
An initiative under active development. This page distinguishes current capability from roadmap — and makes no claim of regulatory approval, deployment or adoption.
The proposition
Most KYC failures are evidence failures. Sentinel is built to catch them before a reviewer signs.
Weak assumptions, missing documents, unsupported conclusions and rationale that would not survive challenge — these are where files fail under scrutiny. Cognitive Sentinel reads a file against a stated evidence and control expectation, surfaces where the support is thin or absent, explains why, and hands a clear, evidence-linked finding to the person accountable for the decision.
How it works
Six principles it is built on
The design goal is control effectiveness with accountability intact — not automation for its own sake.
Evidence-linked findings
Every observation is tied to the specific document, field or gap it rests on — not a black-box score. A reviewer can see exactly what triggered a challenge.
Explainable by design
Each finding carries its reasoning in plain language and against a stated control expectation, so it can be understood, tested and overridden — not just accepted.
Alternative interpretation
Where evidence supports more than one reading, the tool surfaces the alternative rather than forcing a single verdict — leaving judgement where it belongs.
Human review & accountable sign-off
Nothing is decided by the system. Findings are proposed to an accountable reviewer who confirms, amends or rejects — the decision and the sign-off stay human.
Auditability
The basis for each challenge, the reviewer's decision and the rationale are captured as a record — the evidence trail a second line or a regulator expects to see.
Control support, not replacement
Positioned as a method for improving control effectiveness and consistency — support for the reviewer and the second line, not a substitute for either.
Accountability stays human
The system challenges. The human decides.
Cognitive Sentinel is deliberately a decision-support tool. It can argue that evidence is weak; it cannot decide that a customer is acceptable. Accountable judgement — and the regulatory responsibility that comes with it — stays with the reviewer, the second line and the MLRO.
- It does not make or automate regulatory decisions
- It does not replace the accountable human reviewer or mlro
- It does not clear or approve customers on its own
- It does not assert a customer is compliant without human confirmation
Honest about maturity
Current capability, and where it is going
Today Current capability
- Reviews a KYC/CDD file against a stated evidence and control expectation
- Flags weak, missing or unsupported evidence, linked to the source
- Explains the basis for each challenge in plain language
- Proposes findings to a human reviewer for confirmation or override
- Captures the decision and rationale as an auditable record
Roadmap In development
- Deeper coverage across additional entity and structure types
- Configurable control libraries mapped to firm-specific policy
- Integration patterns with case-management and workflow systems
- Expanded second-line QA and sampling support
Roadmap items are planned directions, not delivered features.
Why it is credible
Designed by financial crime practitioners, not model marketers
Cognitive Sentinel encodes the same practitioner judgement behind the rest of the practice — what a defensible file looks like, where evidence typically fails, and what a regulator actually tests. The technology is a method for applying that judgement consistently, with the human firmly in the loop. See the wider AI-enabled compliance approach and the practitioner background behind it.
Speak to the practice
Explore whether Sentinel fits your assurance model.
A confidential conversation about your KYC assurance, second-line QA and where explainable AI could strengthen control effectiveness — with accountability kept where it belongs.