Measuring AI on the work that matters.
Financial crime tasks, reference datasets, and results with the methodology behind them.
AI Research
Advancing AI for
banking operations.
The research behind more capable, reliable AI. Built around the complexity of financial crime and the people who work to prevent it.
Explore our researchSphinx / Applied intelligence
Interpretable Agentic Framework / IAF
Three agents. Every angle.
A case for. A case against. A judgment grounded in your policy. See how Sphinx makes the reasoning visible.
One case. The same facts.
The Prosecutor and Defender receive the same case evidence, with separate instructions and tools. Each builds an independent argument; neither can rewrite the other’s work.
Make both sides of the case.
The Prosecutor builds the risk case. The Defender looks for weak matches, missing context, and mitigating evidence. The disagreement becomes part of the record.
Apply the policy, not an average.
The Judge weighs both arguments against a signed, versioned rubric. It returns a recommendation, flags ambiguity, or escalates the case when the evidence is incomplete.
A recommendation you can inspect.
Your team reviews the evidence, both arguments, and the rubric behind the recommendation. Inputs, outcomes, and any human overrides stay in the case record.
Inside a screening reviewIllustrative case
Shared case fileA name-screening alert
Customer record · Watchlist · Identity documents
Prosecutor
Build the risk case
- Strong name similarity
- Country appears in both records
Defender
Challenge the match
- Date of birth does not match
- Source profile is incomplete
Judge
Weigh both arguments against policy
Identity conflict+Incomplete evidence→Escalate
Ready for human review
Evidence, both arguments, and rubric preserved.
The research agenda
From models to measurable outcomes.
Benchmarks, technical releases, and the methods behind them.
Intelligence built for financial complexity.
Models and agent systems for understanding businesses, investigating risk, and examining documents.
Beyond the right answer.
Evaluating the evidence, reasoning, and adherence to procedures behind every decision.
A closer look at how it works.
Technical whitepapers, experiments, and lessons from building AI for banking operations.
Upcoming research releases will include supporting papers, repositories, and evaluation results.
Our point of view
A score is only useful if you
know what it measures.
Financial crime is full of edge cases. The way we evaluate AI should reflect that.
01
Start with the actual work.
Screening a customer. Tracing ownership. Finding a forged document. Measure the task, not just the response.
02
Make the criteria explicit.
Correctness, supporting evidence, and procedure adherence belong alongside speed and cost.
03
Put results in context.
A useful result comes with its dataset, model version, evaluation method, and limitations.
Continue exploring
The thinking behind the work.
Explore Sphinx’s published guides and perspectives on AI and financial crime.