Products / in development

Sovereign AI products, built for SA regulation.

The global sovereign-AI tools are built for the EU AI Act and HIPAA. None are built for POPIA, SARB, the CMS, or the FSCA. Custos is building the ones that are. Each product below is in active development, and we are taking on a small number of design partners in each industry.

01

Sentinel

Sovereign claims-document AI for medical schemes

POPIA §26 · CMS In development

The problem

Schemes are buried in claims, pre-authorisations, and PMB disputes. An LLM could read and triage them in seconds, but every document is special personal information: it cannot touch a foreign API or leave SA jurisdiction.

What it does

  • + Reads claims, pre-auths, and member correspondence on infrastructure you control
  • + Drafts responses and flags PMB and clinical-coding issues for a human to approve
  • + Keeps every byte of member health data inside SA, with a full audit trail for the CMS

Industry

Medical Aid

Who it's for

Open and restricted schemes and their administrators handling high claim volumes.

View details →
02

Ledger

Sovereign AML and transaction-monitoring copilot

SARB Directive 3 · FIC In development

The problem

AML analysts manually review thousands of flagged transactions. AI could draft the suspicious-transaction narratives, but transaction and KYC data sits under POPIA, FIC, and SARB cloud rules at once. It cannot route through an offshore model.

What it does

  • + Drafts STR and SAR narratives from flagged activity, analyst-in-the-loop
  • + Summarises customer risk and surfaces the reasoning behind every flag
  • + Runs entirely on owned or sovereign infrastructure, explainable to a regulator

Industry

Banking

Who it's for

Retail and business banks running high-volume financial-crime operations.

View details →
03

Adjudicator

Sovereign claims and underwriting explainability engine

SAM · FSCA · TCF In development

The problem

Insurers want AI in claims and underwriting, but SAM, FSCA, and Treating Customers Fairly require every automated decision to be explainable and fair, to a regulator and to a customer. A black-box foreign model cannot satisfy that.

What it does

  • + Produces a plain-language, auditable rationale for every AI-assisted decision
  • + Tests outcomes for fairness and surfaces where a decision needs human review
  • + Defensible to the FSCA, on infrastructure that keeps policyholder data sovereign

Industry

Insurance

Who it's for

Life and short-term insurers putting AI into claims or underwriting.

View details →
04

Custodian

The POPIA control plane for any AI workload

POPIA §72 · cross-sector In development

The problem

Whatever AI you run, you have to prove where personal information went during inference, not just at rest. Almost no one can answer that today. It is the question a regulator or a legal team asks first.

What it does

  • + Sits over any AI workload and maps where personal data moves at every layer
  • + Produces the data-residency and access audit trail legal and the regulator need
  • + Turns "is our AI compliant?" into a document you can hand over, not a guess

Industry

All three

Who it's for

Any regulated enterprise that needs to evidence AI data sovereignty.

View details →

Design partners

Build the product you actually need, with us.

A design partner gets the product shaped around their real workload and regulatory frame, early access as it is built, and pricing that reflects coming in first. We take on a small number per industry. If your team is wrestling with one of these problems now, that is the conversation to have.

Become a design partner