Customer story · DataWhisper
DataWhisper’s agents act autonomously inside regulated workflows.
SmartInsights.CortexOS is integrating with Humanos so relevant legally consequential actions an agent takes are verified, and proved, before execution.
DataWhisper
TL;DR
Legally consequential actions, verified and proved before execution.
SmartInsights.CortexOS is the governance-first agentic operating system for regulated industries. Humanos integrates as its external mandate verifier, branded GuardianShield Consent: the layer that lets a SmartInsights.CortexOS agent prove a human authorized an action, to any party.
Inside SmartInsights.CortexOS, the human-in-the-loop (HITL) engine holds legally consequential actions at an approval gate. For the defined class where the approving party is external, or the proof must be independently verifiable, it calls Humanos.
Relevant legally consequential actions are verified before they commit. The act-or-not boundary is a yes or no, with no LLM in the verification path.
Issue once, verify anywhere.
01 · The Customer
The governance-first agentic OS for regulated industries.
SmartInsights.CortexOS orchestrates multi-agent teams, called Pelotons, inside enterprise workflows: onboarding, dispute resolution, claims, KYC, attestations. Governance is enforced at the infrastructure boundary, not inside a prompt. Every action passes an 18-stage tool gateway and is recorded in a hash-chained, tamper-evident audit trail.
Inside SmartInsights.CortexOS, the human-in-the-loop (HITL) engine holds legally consequential actions at an approval gate. For the defined class where the approving party is external, or the proof must be independently verifiable, it calls Humanos. GuardianShield, the governance and audit plane, records and proves it — branded inside CortexOS as GuardianShield Consent.
Explore SmartInsights.CortexOS at datawhisper.co.uk.
- Channels
- Channel governance
- Agents
- Agentic AI Governance · HITL engine
- Core
- Orchestration · memory · identity
- Roots
- GuardianShield · GuardianShield Consent
- Runtime
- SmartInsights.CortexOS · v5.2
- Integration
- humanos.verify() · one call
02 · Selective by Design
Selective by design, not universal.
GuardianShield Consent does not replace the SmartInsights.CortexOS internal HITL model. Most approvals, escalations and operator overrides resolve inside SmartInsights.CortexOS, with no external dependency and no added latency.
Humanos is invoked only when the proof must survive outside DataWhisper’s systems, or the approving party is external.
If internal governance is sufficient, the external call never happens.
03 · The Integration
One verify() call, at the SmartInsights.CortexOS HITL gate.
Before a legally consequential action commits, the SmartInsights.CortexOS HITL engine calls humanos.verify(). Identity, scope, counterparty, amount and validity are checked against a signed mandate, and a deterministic allow or deny is returned.
No LLM sits in the verification path.
- ✓ Commit
- Resolution commits.
- Within mandate, within constraints, within validity. SmartInsights.CortexOS writes the resolution to the case of record, and a portable proof is attached, verifiable by any party with the right access.
- ⟲ Recover
- Recover, then continue.
- Out of scope, expired or revoked. The HITL engine requests step-up approval in real time, the mandate is updated, and execution resumes once authorization is valid. The agent never silently fails or guesses.
04 · How It Works
Five stages, from mandate to proof.
The stages below walk through one dispute case end-to-end — issued by DataWhisper Operations, verified by Humanos at the SmartInsights.CortexOS HITL engine gate, and committed only after a deterministic yes/no.
- 01
A human authorizes scope.
A DataWhisper Operations Lead authorizes scope, ceiling and validity. Humanos issues a machine-verifiable W3C Verifiable Credential, signed once and reusable across every case.
- 02
SmartInsights.CortexOS prepares the action.
The dispute-resolution Peloton assembles the proposed resolution — refund, fee adjustment, goodwill credit — and attaches the mandate to its outbound action as x-humanos-mandate.
- 03
The HITL engine verifies.
The HITL engine calls humanos.verify() against the mandate. Identity, scope, counterparty, amount and validity are checked in 82 ms — deterministic, no LLM in the path.
- 04
Commit, or step up.
Authorized → SmartInsights.CortexOS commits the resolution. Out of scope — a £45,000 settlement against a £25,000 ceiling — the gate blocks and triggers a real-time step-up.
- 05
Anchor the proof.
Each authorized action emits a cryptographic proof, recorded in both the SmartInsights.CortexOS tamper-evident audit trail and the independent consent record. Auditors, regulators and partners verify it directly, without reconstructing internal logs.
05 · What It Delivers
Provable authorization, by default.
- Verified
- Verified before execution.
- Relevant legally consequential actions are verified before they commit. The verify call is inline at the HITL gate, not after the agent has acted.
- Deterministic
- Probabilistic AI, deterministic answer.
- Agents stay probabilistic in reasoning. The act-or-not boundary is a yes or no, with no LLM in the verification path.
- External reach
- Human-in-the-loop beyond the operator.
- Approval and consent can come from external counterparties and data subjects, not only internal operators. The loop extends past DataWhisper’s boundary.
- Portable
- Authorization travels with the action.
- Issue once, verify anywhere. The proof is recorded in the SmartInsights.CortexOS tamper-evident audit trail, lives inside GuardianShield Governance, and is verifiable by any permitted party.
06 · Where It Applies
Anywhere a regulated agent acts on a human’s authority.
- Engage & Grow · Fraud & Risk
- Disputes & chargebacks
- Refunds, settlements and goodwill credits, each authorized at the gate and carrying a portable proof.
- Engage & Grow · Fraud & Risk
- High-value payment execution
- Above an operator-defined threshold, cryptographic proof that a human authorized execution at a specific amount.
- AI Onboarding
- KYB / KYC & sanctions
- A signed human authorization record provides audit-ready evidence for FCA or equivalent review.
- Engage & Grow
- Contract & terms acceptance
- Tamper-proof evidence that a counterparty accepted terms at a specific time and scope.