iShield Solutions Series, Platform & AI · AI Platform
Sovereign AI Gateway & Agent Governance
Every Model Call Through One Policy Plane, Every Agent Action on the Record
Sovereign AI Gateway & Agent Governance routes every model call and agent action through one place where residency, redaction, cost and approval policies apply, and keeps the policies, exceptions and evidence that governance bodies need.
Governed LLM access, agent orchestration, prompt firewall and human-in-the-loop approvals.
Who leads, who contributes
- LeadiShield Link
- ContributoriShield Nexus
- ContributoriShieldGRC · iShield GRC Ltd.
Provider brokering, prompt firewall and PII redaction, residency/consent/cost policy, agent orchestration with tool approvals, and the evaluation harness.
The enterprise challenge
AI adoption moves faster than AI governance.
Teams connect models to business data through many paths: chat tools, embedded copilots, scripts and agents. Each path makes its own choices about where prompts go, what data they carry and which tools an agent may call. Governance learns about them from expense reports.
A single policy plane changes that. When every call passes one gateway, policy is applied consistently, agent actions that touch business systems wait for approval, and the record of what happened exists by default rather than by request.
Where iShield Nexus contributes
iShield Nexus contributes the governance record — policies, exceptions and audit logs in Risk & Compliance. iShield Link leads the gateway, prompt firewall, policy engine and agent orchestration; iShieldGRC is listed for AI usage assurance.
The Nexus angle
AI usage needs recorded policies, exceptions and audit evidence.
- Technology services and controls are managed apart from the business process
- Evidence and ownership are fragmented
Solution overview. One control plane for every LLM and agent in the enterprise. AI Gateway brokers OpenAI, Anthropic, Azure OpenAI, Mistral, on-prem models with prompt-firewall, PII redaction, policy enforcement, cost caps and full audit. Agent Orchestration runs multi-step agents with tool approvals.
Buyer challenges
Shadow model use
Prompts carrying business data reach external providers through tools nobody registered, in regions nobody chose.
Agents with open hands
Agents can call business tools directly, so an automated action and an approved action look the same in the logs.
Policy without evidence
AI policies are approved by committees, but there is no record showing they were applied to a given call.
Cost without control
Model spend grows by team and provider with no shared limits or attribution.
Decision owners
- AI governance board
- Which AI use cases are permitted, under which policy and risk classification.
- Platform owner
- Which providers and models are available, in which regions, at what cost limits.
- Business process owner
- Which agent tool calls need human approval before they run.
- Risk owner
- Whether a policy exception is accepted, and for how long.
Who leads, who contributes, and where each stops
This is a cross-product solution. Each platform keeps its own responsibility and boundary; none is presented as a module of another.
- iShield LinkLead
Owns
Provider brokering, prompt firewall and PII redaction, residency/consent/cost policy, agent orchestration with tool approvals, and the evaluation harness.
Boundary
Applies the policies the organization sets; it does not decide which use cases are acceptable.
- iShield NexusContributor
Owns
The governance record: AI usage policies, exceptions with expiry and audit logs held in Risk & Compliance.
Boundary
Does not broker model calls or run agents.
- iShieldGRC (iShield GRC Ltd.)Contributor
Owns
AI usage assurance through Audit Management, as listed in the composition.
Boundary
A separate related offering of iShield GRC
Module responsibilities
| Module | Platform | Responsibility in this solution | Hands on |
|---|
Published composition, by owning platform
The modules in the solution's published composition, grouped by the platform that owns them.
iShield Link
iShield Nexus
iShieldGRC
Why this composition
Provider broker, prompt firewall, agent orchestration and audit must share one policy plane or governance is theatre. Adding iShieldGRC ties every model call back to EU AI Act and NCA AI evidence requirements.
Cross-platform journey
From a prompt to an approved, evidenced action
Follow one agent request through the policy plane. Applications and agents enter on the left, providers sit on the right, and governance sits above the plane, receiving the record of every decision.
- Register the use case · Risk & Compliance iShield Nexus
- Request · AI Gateway iShield Link
- Screen · Prompt Firewall & PII Redactor iShield Link
- Apply policy · Policy Engine iShield Link
- Approve the tool call · Agent Orchestration iShield Link
- Evaluate · Eval & Red-team Harness iShield Link
- Decide exceptions · Risk & Compliance iShield Nexus
- Assure · Audit Management (AI usage) iShieldGRC
Stage 1 of 8 · iShield Nexus Register the use case
The AI use case is recorded with its policy, owner and risk classification.
- Decision
- The AI governance board approves the use case.
- Exception path
- An unregistered use case is blocked at the gateway until it is recorded.
- Recorded as
- Policy and use-case record
Next: Request in AI Gateway (iShield Link) — the work passes to another platform
This journey shows how responsibilities are intended to divide between the platforms. It describes an operating model, not a pre-built integration between them.
Decisions, exceptions and who owns them
Policies and exceptions are owned and reviewed.
| Decision | Accountable owner | Evidence recorded | Exception path |
|---|---|---|---|
| Permit an AI use case | AI governance board | Use-case record with policy and risk classification | Use cases outside policy are refused or accepted as time-bound exceptions. |
| Make a provider or model available | Platform owner | Availability, region and cost settings | Calls to unavailable models are refused with the reason. |
| Approve an agent's tool call | Business process owner | Approval record linked to the tool call | Declined calls return to the agent; nothing runs without approval where policy requires it. |
| Accept a policy exception | Risk owner | Exception with compensating measure and expiry | Expired exceptions return for decision. |
How AI obligations are mapped
AI policies and their risk classifications are written and approved by the organization's AI governance board and held in Nexus Risk & Compliance. The gateway applies those policies; neither platform ships a pre-built interpretation of an AI regulation.
- Each policy names its owner, the use cases it covers and the regulation it responds to.
- Gateway decisions reference the policy version that produced them.
Regulatory context
Regulations and frameworks organizations in this domain commonly work with. Listed as context for planning, not as a certification, attestation or pre-built mapping.
Intelligence in explicit modes, by platform
Nexus AI capabilities assist the people involved; decisions remain with accountable owners.
- Assistance drafts, answers or summarizes
- Recommendation proposes for a person to accept
- Predictive anticipates or flags
- Workflow-triggered runs only as configured and approved
Evidence mapping
RecommendationSuggests gateway and exception records as evidence for AI governance controls.
Risk & Compliance — Evidence mapping
Recognized practice behind the operating model
Mappings describe how the operating model aligns with recognized practice. They are not attestations or certifications.
The governance record follows the risk-management process of ISO 31000 mapped on the accepted Risk & Compliance page.
| Framework | Practice | How the solution operates | Relationship |
|---|---|---|---|
| ISO 31000:2018 | Risk treatment, monitoring and review | AI use cases and policy exceptions are owned, treated and reviewed with expiry.Mapped on the accepted Risk & Compliance page. | Compliance-supporting capability |
What changes for the organization
Outcomes are described qualitatively. Figures are published only once evidenced.
AI use you can see
Every registered use case and every call passes one place, with its policy on the record.
Agents that ask first
Tool calls that touch business systems wait for approval where policy requires it.
Policy with proof
Each gateway decision names the policy version that produced it.
Spend under owner control
Cost limits are set by owners and overruns wait for a decision.
What sets this solution apart
Governance kept apart from execution
The governance record lives in Nexus Risk & Compliance, separate from the gateway that executes calls.
Delivery and industry context
Planning information carried over from the solution record.
30 / 60 / 90-day value plan
Days 0–30 — Foundation
Stand up the iShield workspace, integrate sources of truth, ingest baseline data.
- Tenant provisioned in chosen region (KSA / UAE / EU / US)
- SSO + RBAC mapped to enterprise identity (Entra ID / Okta)
- Top 5 integrations connected and reconciled
- Baseline KPIs measured before any change
Days 31–60 — Activation
Turn on automation, AI agents and guardrails for the priority value stream.
- Workflows live for the top 3 use cases
- Link AI agents enabled with explicit guardrails and approvals
- Power-user enablement and admin training complete
- First measurable KPI improvement vs baseline
Days 61–90 — Scale
Industrialize, roll out to additional teams, prove ROI to the steering committee.
- Solution rolled out to additional business units / geos
- Executive dashboard with target vs actual outcomes
- Continuous improvement backlog established
- Signed ROI memo and renewal / expansion plan
Personas served
Illustrative industry scenarios
Representative situations by industry. Regulations named here are context, not certifications or commitments.
Banking
All LLM calls brokered with PII redaction, cost caps per LoB, eval evidence pack for SAMA AI guidance and EU AI Act high-risk classification.
Public Sector (KSA)
NCA AI Ethics-aligned agent orchestration, residency-pinned models in KSA, citizen-data prompt firewall.
Healthcare
Clinical-decision-support agents with mandatory human approval, PHI redaction on every prompt, full audit per regulator request.
Telecom & ISVs
Per-tenant LLM brokering, cost and rate-limit guardrails, eval/red-team harness on customer-facing copilots.
Related Platform & AI solutions
API, Integration & Event Platform
API gateway, iPaaS, event bus, connectors and EDI on one sovereign runtime.
OpenMulti-Tenant SaaS Enablement
Tenant lifecycle, isolation, metering, entitlements and per-tenant residency for SaaS builders.
OpenSovereign Data Fabric & Master Data
Federated query, lineage, consent and golden records — without copying data across borders.
Open
