Production AI Institute · PSF v1.1 open standard

AI Right-To-Know AI Data Use Index Check My AI Tools Policy Change Watch Agent Readiness Public Benchmark Contact

Official Document · PAI-BP-001

AIDA Exam Blueprint

AI Deployment Associate — Official domain weightings, topic areas, and format specifications. Updated Q1 2026.

Exam Format

Questions

45 min

Time Limit

75% (15/20)

Pass Mark

72 hours

Retake Wait

📋 Question format: Multiple choice, single best answer

🌐 Delivery: Online, open-book, unproctored

🎓 Prerequisites: None — open to all

Domain Weightings

PSF-1

Input Governance20%

PSF-2

Output Validation20%

PSF-3

Data Protection10%

PSF-4

Observability15%

PSF-5

Deployment Safety12%

PSF-6

Human Oversight15%

PSF-7

Security5%

PSF-8

Vendor Resilience3%

Weightings represent approximate percentage of exam questions drawn from each domain. Actual question counts vary by exam sitting.

Domain Detail

PSF-1

Input Governance

20% of exam

—Prompt injection taxonomy (direct, indirect, multi-turn)
—Input validation architectures (allow-list vs deny-list vs semantic)
—Jailbreak pattern recognition and mitigation
—Multi-modal input risks (image, audio, document injection)
—Rate limiting and abuse detection patterns
—Input logging and audit trail design

PSF-2

Output Validation

20% of exam

—Output validation layers and post-processing pipelines
—Hallucination detection and confidence calibration
—PII/sensitive data detection and redaction in outputs
—Content policy enforcement architecture
—Structured vs unstructured output risk profiles
—Output logging, sampling, and quality monitoring

PSF-3

Data Protection

10% of exam

—Training data lineage and contamination risks
—GDPR/CCPA implications for AI-processed data
—Data retention policies for AI inputs and outputs
—Cross-border data transfer considerations
—Model memorization and data extraction attacks
—Consent management for AI feature usage

PSF-4

Observability

15% of exam

—AI-specific metrics: latency p50/p95/p99, token throughput, error rates
—Quality drift detection and statistical process control
—Alerting thresholds and escalation paths
—Tracing AI call chains in distributed systems
—Cost monitoring and token budget management
—Human review queue sampling telemetry

PSF-5

Deployment Safety

12% of exam

—Staged rollout, shadow mode, and canary deployment patterns
—Rollback procedures for model and prompt updates
—AI incident classification and severity levels
—Post-incident review process for AI failures
—Kill switch architecture and emergency stop procedures

PSF-6

Human Oversight

15% of exam

—Human-in-the-loop design patterns and when to require them
—Confidence thresholds for autonomous vs assisted operation
—Override mechanisms and graceful degradation
—Feedback loop design for continuous improvement
—Operator vs user trust levels in LLM deployments

PSF-7

Security

5% of exam

—Principle of least privilege for AI agents
—API key management, rotation, and secret hygiene
—Role-based access control for AI features
—OAuth scopes and delegated permission models
—Multi-tenant isolation in shared AI infrastructure
—Audit logging for access events

PSF-8

Vendor Resilience

3% of exam

—Provider SLA evaluation and gap analysis
—Multi-provider fallback architecture
—Model deprecation risk and migration planning
—Total cost of ownership across provider tiers
—Contractual data handling obligations

PSF Deep Dive Guides

Complete implementation guides for every exam domain. Each guide covers the theory tested in the exam plus real production patterns.

D1 Input GovernancePrompt injection, schema validation, input contracts D2 Output ValidationThree-layer output contract, confidence thresholds D3 Data ProtectionPII masking, retention, why no framework covers it D4 ObservabilityLogging schema, essential alerts, drift detection D5 Deployment SafetyModel versioning, canary releases, rollback D6 Human OversightHITL patterns, autonomy taxonomy, GDPR Art 22 D7 SecurityAI threat model, prompt injection, supply chain D8 Vendor ResilienceLock-in taxonomy, deprecation playbook, exit strategy

Certificate Maintenance

Validity Period

AIDA certification is valid for 2 years from the date of issue. Renewal requires passing a 20-question recertification exam covering updates to the PAI Production Safety Framework.

Version Alignment

The AIDA exam is aligned to PSF v1.1. Major PSF version changes trigger a 6-month transition period during which both exam versions are accepted. Certificates note the PSF version in force at issue; v1.0 remains valid where so stated.

CAIA — Certified AI Auditor

PAI-8 Audit Track Exam Blueprint

Audit Track · 30 questions · 60 minutes · Pass threshold: 22/30 (73%) · Fee: $97

Take Exam →

C1AI Governance

12%

·Governance accountability models
·Policy vs operational governance
·Decision gate design
·AI use case registry
·RACI for AI risk

C2Risk Assessment

13%

·Risk tiering criteria
·Trigger-based reassessment
·Third-party AI risk inclusion
·High-risk use case classification
·Risk register design

C3Data Stewardship

12%

·Training data provenance
·Consent and lawful basis for training
·PII handling in AI pipelines
·Data lineage documentation
·Retention and deletion for training artefacts

C4Model Validation

13%

·Pre-deployment evaluation design
·Bias and fairness testing
·Independent review requirements
·Benchmark selection for production populations
·Post-deployment drift monitoring

C5Human Oversight

12%

·Autonomy scope definition
·Override mechanism design
·Escalation path requirements
·Staff awareness and training
·Override logging and audit

C6Incident Response

12%

·AI incident classification taxonomy
·Severity tiers for AI-specific harms
·Response SLA design for AI incidents
·Post-incident review requirements
·Regulatory notification obligations

C7Audit Trail

13%

·Decision-level logging schema
·Log retention aligned to challenge windows
·Explainability artefact requirements
·Immutable audit trail design
·Reconstruction requirements for contested decisions

C8Vendor & Supply Chain

13%

·Third-party AI inventory
·Model deprecation risk and contractual protections
·Vendor SLA design for AI continuity
·Alternative model evaluation
·Supply chain risk assessment

Exam format and approach

Question type

All questions are scenario-based vignettes — realistic organisational situations requiring the candidate to identify the correct maturity classification, finding type, or evidence requirement.

Maturity model

Questions test knowledge of PAI-8 L0–L3 maturity levels. Candidates must distinguish between control absence (L0), documented-but-inactive (L1), actively managed (L2), and continuously improved (L3).

Prerequisites

CAIG certification is recommended but not required. Candidates should be familiar with the PSF 8 domains (AIMA level). The exam assumes AI governance fluency and production AI deployment context.

PAI-8 Maturity Model Reference

L0 — Absent

No controls. AI deployed without governance, assessment, or documentation.

L1 — Basic

Controls documented but not consistently operational or evidenced.

L2 — Managed

Controls operational with documented process, evidence, and regular cadence.

L3 — Optimised

Controls continuously improved with metrics, benchmarking, and governance review.

CAIA Exam Preparation Resources

Study materials aligned to each PAI-8 control. All resources are free.

PAI-8 AI Safety StandardThe definitive reference — all 8 controls, maturity levels, and indicators CAIA Audit Case Studies8 audit vignettes — one per PAI-8 control with findings and evidence requirements The Agent OperatorContext for who gets audited and why AI governance credentials matter PAI-8 Assessment ProcessHow real assessments are conducted — mirrors the exam scenarios D1 Input Governance (PSF)Underlying technical controls mapped to C2 and C4 audit requirements D6 Human Oversight (PSF)HITL patterns and autonomy taxonomy — directly maps to C5 exam questions

Production AI Institute · Blueprint version 1.0 · Effective January 2026

Questions about exam content? Contact us via the contact form