Classify
Assign a risk tier per workload from impact, autonomy and data sensitivity.
Done when: The tier is recorded before any model touches production data.
AI Assurance Framework (TRUST)
AI Assurance Framework
One control set that answers NIST, ISO 42001, the EU AI Act and the Saudi regulators at once.
The business problem
Every regulator, a new project. Every audit, a scramble.
Controls are written separately for each standard, evidence is collected by hand, and AI launches wait for approval.
What you hear inside the organisation
Can we prove how this model made the decision?
Compliance blocked the launch again.
We answered the same audit question three times.
Who approved this AI use?
Who it serves and what changes
Your teams, before and after the AI Assurance Framework
Client teams and what they need
Risk and compliance
One control set, mapped once
Internal audit
Evidence that is collected automatically
Legal and privacy
Clear PDPL positions
AI product teams
Faster, predictable approvals
The outcome you get
- One control library mapped to NIST, ISO 42001 and Saudi rules
- Risk tiering for every AI use
- Evidence captured as work happens
- Approvals that no longer stall delivery
A worked example
An insurer launching AI claims triage
The problem we saw
Compliance asked for evidence against several standards. Each review started from scratch and the launch kept slipping.
How the framework solved it
We tiered the use case by risk, applied one control set mapped to every standard, and captured evidence inside the delivery work.
The business result
Compliance signed off once, auditors read the same evidence pack, and the next AI use case followed the same path faster.
Illustrative example based on common client situations.
In short
What is the AI Assurance Framework?
AI Assurance Framework (TRUST): One control set that answers NIST, ISO 42001, the EU AI Act and the Saudi regulators at once.
Assurance fails when each regulator triggers a separate project.
- How it moves
- Each step closes with a check before the next begins
- Owned by
- Owned by risk and compliance, operated with the delivery team that builds each workload.
- You keep
- Tiered control matrix, Audit evidence pack, Standards crosswalk
How it works
How the AI Assurance Framework works
The same evidence pack serves an internal audit, a regulator request and a customer security review. It is produced once.
Control
Apply the control set that the tier requires, no more, no less.
Done when: Every required control has an owner and a test.
Evidence
Capture logs, evaluations, approvals and data lineage as the release happens.
Done when: Evidence is produced by the pipeline, not assembled before an audit.
Attest
Sign off per release against the mapped clauses, with named approvers.
Done when: No production release without a current attestation.
Monitor
Watch drift, incidents and control failures, and re-tier when behaviour changes.
Done when: A failed control raises an incident with a defined response time.
Is it right for you?
Is the AI Assurance Framework right for you?
A strong fit if
- You operate in a regulated sector
- AI launches wait on compliance reviews
- You expect an AI audit or certification
Probably not yet if
- AI use is internal, low risk and experimental
- You need legal advice rather than an operating control set
Free assessment
Discuss the AI Assurance Framework with our team
A working session with our senior team. We review what fits your situation, what does not, and where to start.
A focused conversation
We listen to your situation and goals.
Fit review
Which parts of this framework apply to you, and which do not.
A written view
Clear next steps, whether or not you work with us.
Go deeper
Applying the AI Assurance Framework in practice
Open only the part you need: how it starts, what your team keeps, maturity levels, sector rules and the evidence a reviewer will ask for.
Assurance fails when each regulator triggers a separate project. TRUST defines controls once, then maps the same evidence to every framework a reviewer might ask about.
Version
v2.2
Last reviewed
August 2026
Family
AI Operating System
How this was developed: Built from delivery practice on Saudi and international engagements, then reconciled clause by clause against the published standards below. Revised when a standard is updated or a gate proves unworkable in the field.
One-pager
Download the TRUST one-pager
Stages, exit gates, artefacts, standards crosswalk and FAQs for AI Assurance Framework on one printable page, share it with the committee.
- Week 1
Tier every live and planned AI use by impact, and write the reasoning down.
- Week 2
Map controls to the clauses your reviewer cites, and mark honestly what does not exist yet.
- Week 3
Assemble the evidence pack so a reviewer can read it without a workshop.
- Week 4
Stand up monitoring with thresholds and a named escalation path.
- End of the first thirty days
Tiered control matrix
Controls per risk tier, each with an owner and a test.
Audit evidence pack
Release-time evidence mapped to every referenced clause.
Standards crosswalk
One table showing which control answers which framework.
L1
Undocumented
AI use is known informally; nothing is tiered.
L2
Policy-level
Policies exist; evidence does not.
L3
Evidenced
Each control produces an artefact on a schedule.
L4
Monitored
Drift, incidents and overrides are reviewed by a committee.
Banking & finance
Model risk management and SAMA expectations are folded into the tiering rules.
- Who reviews it
- SAMA, plus PDPL and NCA controls
- Systems in scope
- Core banking, cards, KYC and AML systems, credit bureau feeds, the general ledger
- Where it lands first
- KYC and AML review, credit decisioning support, complaint and dispute handling
1
Classify
- Risk tier record
- Impact assessment
2
Control
- Control matrix
- Control tests
3
Evidence
- Evidence pack
- Model card
4
Attest
- Attestation record
- Clause crosswalk
5
Monitor
- Drift and incident log
- Re-tier decision
The same evidence pack serves an internal audit, a regulator request and a customer security review. It is produced once.
Stage
Exit gate
Artefacts
1
Classify
Assign a risk tier per workload from impact, autonomy and data sensitivity.
- Signed off by
- Risk and compliance lead
- Typical duration
- Scoped with you
Exit gate
The tier is recorded before any model touches production data.
Evidence: Risk tier decision recorded with the reasoning
Artefacts
- Risk tier record
- Impact assessment
2
Control
Apply the control set that the tier requires, no more, no less.
- Signed off by
- Control owners per domain
- Typical duration
- Scoped with you
Exit gate
Every required control has an owner and a test.
Evidence: Control set mapped to the clause it answers
Artefacts
- Control matrix
- Control tests
3
Evidence
Capture logs, evaluations, approvals and data lineage as the release happens.
- Signed off by
- Internal audit or the second line
- Typical duration
- Scoped with you
Exit gate
Evidence is produced by the pipeline, not assembled before an audit.
Evidence: Evidence pack that survives review without a workshop
Artefacts
- Evidence pack
- Model card
4
Attest
Sign off per release against the mapped clauses, with named approvers.
- Signed off by
- Executive risk owner
- Typical duration
- Scoped with you
Exit gate
No production release without a current attestation.
Evidence: Signed attestation naming the accountable executive per tiered use
Artefacts
- Attestation record
- Clause crosswalk
5
Monitor
Watch drift, incidents and control failures, and re-tier when behaviour changes.
- Signed off by
- Risk committee
- Typical duration
- Ongoing
Exit gate
A failed control raises an incident with a defined response time.
Evidence: Monitoring plan with thresholds and an escalation path
Artefacts
- Drift and incident log
- Re-tier decision
Standard
What this framework satisfies
NIST AI RMFGOVERN · MAP · MEASURE · MANAGE (all four functions)
Govern, Map, Measure and Manage functions map to Classify, Control, Evidence and Monitor.
ISO/IEC 42001Cl. 4–10 and Annex A controls
AI management system clauses 4-10 and Annex A controls.
EU AI ActArt. 6 classification · Art. 9 risk management · Art. 11 technical documentation · Art. 72 post-market monitoring
Risk tiering, technical documentation and post-market monitoring duties.
SDAIA AI EthicsFairness · Transparency · Accountability · Human oversight
Fairness, transparency, accountability and human oversight principles.
PDPL / NDMOPDPL Art. 12 privacy policy · Art. 29 transfer · NDMO personal data controls
Personal data handling, minimisation and cross-border rules per workload.
NCA ECCECC 2-2 identity · 2-12 logging · 2-8 key management · 1-6 change control
Identity, logging, key management and change control for AI systems.
Audit-time archaeology
Reconstructing evidence after the fact costs more than generating it during release.
One control set for everything
Applying high-risk controls to every workload stalls low-risk work and breeds workarounds.
Policy without tests
A control that is not tested is a sentence, not a safeguard.
Built on
- 1
NIST AI Risk Management Framework 1.0 and Generative AI Profile
National Institute of Standards and Technology, Govern, Map, Measure, Manage.
- 2
ISO/IEC 42001: AI management systems
International Organization for Standardization, management system requirements for AI.
- 3
ISO/IEC 23894: AI risk management
Guidance on managing risk across the AI lifecycle.
- 4
EU AI Act
Risk tiering, documentation and transparency obligations.
- 5
SDAIA AI Ethics Principles
Saudi Data & AI Authority, national principles for AI.
- 6
NDMO data management and personal data protection standards
National Data Management Office specifications and domains.
- 7
Personal Data Protection Law (PDPL) and its implementing regulations
Lawful basis, subject rights, transfer and retention.
- 8
NCA Essential and Cloud Cybersecurity Controls
National Cybersecurity Authority control catalogues.
Citations describe the parts of each standard the framework answers. Bilytica is not an accreditation or certification body.
In Saudi Arabia
What this family has to satisfy before a release in Saudi Arabia
The AI operating system frameworks carry a workload from a readiness score to a governed release. These are the Saudi checks each stage is built against.
AuthorityWhat it requiresWhat we buildWhat you keep
SDAIA, PDPL
Purpose, lawful basis and disclosure recorded before the first personal-data field is used.
Diagnosis and assurance stages block a build that has no recorded basis.
Processing register entry attached to the release record.
NCA ECC
Authorisation and containment for anything the system can do on its own.
Agent operations defines action classes, dual control and a tested rollback path.
Containment test result and an authority matrix signed by the business owner.
NDMO data management controls
No workload builds on an uncertified data domain.
The data foundation gate certifies ownership, lineage and quality thresholds first.
Signed domain certificate referenced by the release.
Where it runs
Training, evaluation and serving run inside Saudi Arabia on your infrastructure, with any hosted component documented as a boundary.
Arabic, as it is spoken here
Arabic quality is a release condition: evaluation sets include Saudi Arabic content, not translated English.
FAQ
Frequently Asked Questions
Do we need ISO 42001 certification to use this?
No. The framework is structured so the same controls and evidence would support a certification effort if you later choose one.
How are risk tiers decided?
From three inputs: impact on a person or asset, the degree of autonomy granted, and the sensitivity of the data involved.
How long does it take to stand up the control set?
Timeline agreed with you for the control set itself; after that, evidence accrues automatically with each release rather than being reconstructed at audit time.
Does one control set apply to every AI system?
No. The Classify stage assigns a risk class first, and controls scale with that class, so a low-risk assistant is not governed like a decisioning system.
Which regulations does the evidence serve?
The same artefact set is mapped to NIST AI RMF, ISO/IEC 42001, the EU AI Act, SDAIA AI Ethics, PDPL and NDMO, and NCA ECC, one evidence pack, several reviewers.
What does attestation mean here?
A named owner signs that the declared controls were tested for this release, with the test results attached. Monitoring then watches for drift between attestations.
Where this sits on the spine
Where this connects
Rest of the library
DFFData Foundation FrameworkAIRMAI Readiness ModelVALUEAI Value FrameworkDELIVERAI Delivery FrameworkAGENTAgent Operations FrameworkSOVEREIGNSovereign AI FrameworkDMMData Management Maturity ModelDPAData Platform Architecture FrameworkDQFData Quality & Observability FrameworkDECIDEAnalytics & Decision FrameworkDGOVData Governance Operating ModelSDCFSaudi Data Compliance FrameworkAQSArabic Quality StandardKSA-READYIn-country Deployment Framework
Topic cluster
Explore the topic cluster
Curated links across our Saudi hubs, sector pages and long-form blog research.