Policy center

The rules the marketplace runs on, PDPPL, the national AI ethics guidelines, data classification, and model lifecycle duties, with the platform control that enforces each one.

Assets classifiedin force

100%

every catalog asset carries a PDPPL tier

RAI checklists completeaction open

96%

4 Limited-risk models pending re-attestation

DPIAs on filein force

12/12

all High-risk integrations covered

Audit coveragein force

24 months

immutable, queryable, entity-scoped

Sovereignty by construction

Residency is not a clause in a contract here, governed data, the compute that trains on it, and the inference served from it all sit inside the same national perimeter, and the platform has no path that crosses it.

Governed data, sovereign compute and inference sit inside Qatar's sovereign boundary; the public internet and offshore model APIs are blocked at the perimeter.STATE OF QATAR · SOVEREIGN BOUNDARYدولة قطرGoverned dataMinistry datasets · classified at sourceSovereign computeIn-country GPU clusterInference & agentsServed inside the perimeterEgress-controlled · nothing leaves the perimeterPublic internetopen networkno ingressOffshore model APIshosted abroadno egress
  • No egress from sandboxes
  • Weights and data never leave in-country storage
  • Derived models inherit the strictest upstream classification
  • Every access purpose-bound and audited

Policies in force

Each policy is a live constraint in the publishing and access paths, not a document filed elsewhere.

PDPPL Alignment

Law No. 13 of 2016 · Personal Data Privacy Protection

Every asset carries a PDPPL classification (Public / Government-Restricted / Entity-Private) assigned at publish time and re-validated on each version. Access to restricted assets requires a stated purpose, a bounded duration, and entity-level accountability. Purpose limitation is enforced through entitlements: a grant covers the declared use only, and re-use for a new purpose requires a new request.

  • Classification assigned at publish, audited quarterly
  • Purpose-bound, time-bound entitlements
  • PII scanning on every dataset ingestion
  • DPIA mandatory for High-risk model integrations

National AI Ethics Guidelines

Qatar National AI Strategy · MCIT governance framework

Models are risk-tiered (Minimal / Limited / High) following the national AI governance framework. High-risk models, meaning those affecting individuals' rights, safety, or access to services, require a completed Responsible AI checklist, documented human oversight, and periodic bias re-assessment. The platform's Responsible AI gate blocks publication until these artifacts exist.

  • Risk tiers assigned via structured questionnaire
  • Human-in-the-loop mandatory for High-risk inference
  • Bias assessments re-run on every retrain
  • Explainability method documented per model

Data Classification Policy

National Data Classification Scheme

Datasets follow the national data classification scheme. Government-Restricted data never leaves sovereign infrastructure; Entity-Private data is visible only to the owning entity and named grantees. Derived artifacts (models, embeddings) inherit the strictest classification of their training data unless a documented de-identification review lowers it.

  • Classification inheritance for derived models
  • Sovereign-only storage and compute for restricted tiers
  • Watermarked, logged downloads
  • Annual declassification review cycle

Model Lifecycle & Retirement Policy

TASMU platform standard · MCIT Platform Directorate

A published model is a standing commitment to its consumers, not a one-off deliverable. Owners carry versioning duties for as long as the asset serves traffic: semantic versions with a changelog, a retrain SLA triggered by measured drift, and a fixed deprecation notice period before an endpoint is withdrawn. Retirement is a governed decision with a named migration target, an asset may not simply go dark on the entities that built on it.

  • Semantic versioning with a published changelog per release
  • Retrain SLA: proposal within 10 working days of PSI > 0.25, shipped within 45
  • Deprecation notice: 90 days for Production endpoints, 30 for Staging
  • Consumer migration obligation: a superseding asset and guide before withdrawal

Controls in force

Every policy duty maps to a mechanism and a screen. 7 of 10 run unattended, the rest hold a deliberate human checkpoint.

Automation console
DutyPolicyEnforcing controlVisible in productMode
Purpose limitationPDPPL AlignmentGated access request with a stated purpose and bounded durationAsset card · Access monitoringhuman gate
Entitlement expiryPDPPL AlignmentRenewal nudge at 30 days, endpoint revocation on lapseEntitlement expiry panel · Automation ruleautomated
PII detectionData Classification PolicyPII scan on every ingestion wave, publisher sign-off on any hitPublishing pipeline · Dataset health scoreautomated
Classification inheritanceData Classification PolicyDerived asset inherits the strictest upstream tier and gates itselfModel card lineage · Automation ruleautomated
Sovereign residencyData Classification PolicyEgress scanner blocks and quarantines off-platform transfersAccess monitoring · flagged eventsautomated
Responsible AI attestationNational AI Ethics GuidelinesRAI checklist gate: publication blocked until artifacts existPublishing pipeline stage 3human gate
Bias re-assessmentNational AI Ethics GuidelinesBias suite re-run on every retrain, results attached to the versionModel card metrics · MLOps pipelineautomated
Drift & retrain SLAModel Lifecycle & Retirement PolicyWeekly PSI job opens a pre-filled retrain proposal for the ownerMLOps drift posture · Automation ruleautomated
Deprecation & migrationModel Lifecycle & Retirement PolicyRetirement review with a named migration target and consumer noticeGovernance · Retirement tabhuman gate
Accountability recordAll policiesImmutable audit entry for every access, publish, and governance decisionGovernance · Audit logautomated

Enforcement is automated where it can be

Policy that depends on someone remembering to apply it is not policy. Classification inheritance, PII scanning, egress blocking, entitlement expiry, and drift-triggered retrain proposals all run unattended on the platform: 7 of 10 controls on this page fire without a human in the loop, and every firing writes to the audit trail. The rules themselves are inspectable and toggleable in the Automation console.

What stays human is deliberate, not residual: approving access to restricted microdata, signing off a Responsible AI attestation, and deciding to retire an asset other entities depend on. Those carry accountability that cannot be delegated to a rule, so the platform routes them to a named officer at the owning entity and to the MCIT governance desk together.