IAPP AIGP Course · Module 7 · Summary Sheet p.432

Governing AI Deployment — Release, Monitor, Maintain

The complete Module 7 curriculum: Lesson 1 — key considerations in planning for AI deployment (own model vs third-party); Lesson 2 — governing the release, monitoring and maintenance of the AI model. Every topic from the sheet, with plain-words examples.

2 lessons 3 guardrail tiers 7 proprietary risks 6 failure attributes 16 flip cards
PLAN & DESIGN DATA COLLECTION & PREP MODEL DEV MODEL TEST & EVALUATE DEPLOY MONITOR & MAINTAIN DECOMMISSION
Curriculum map: Lesson 1 (L1) = key considerations in planning for AI deployment — deploying the AI, updating policies, deployment requirements (environment, agentic architecture), proprietary vs third-party trade-offs, and vendor/open-source agreement terms. Lesson 2 (L2) = governing the release, monitoring and maintenance of the AI model — periodic assessment, performance monitoring, incident response, and the downstream-consequence guidelines. This module follows the sheet exactly — nothing added, nothing dropped.
Scroll to begin ↓
1
Curriculum block 1 · Lesson 1 (L1)

Planning for AI Deployment

Before anything ships: who built it, who uses it, are your policies ready, and where does it run? The sheet's L1 planning considerations, with plain-words examples.

1.1 Deploying the AI

Who developed it?

In-house team, third-party vendor, open-source community, or a foundation model provider?

Example: a bank deploys a credit model built in-house, but the LLM chatbot is from a vendor — different provenance, different accountability.

Is it customised / customisable?

Off-the-shelf vs fine-tuned vs fully custom — determines how much control you have over behaviour, bias, and security.

Example: a vendor chatbot with no fine-tuning access can't be re-trained to remove a discovered bias.

Who will use it?

Employees, customers, or the public — each audience changes the risk profile and disclosure duties.

Example: an internal copilot for staff vs a public-facing claims assistant have very different oversight needs.

1.2 Evaluate & update existing privacy, security policies

Exam anchor — the readiness assessmentA readiness assessment is the gate between "built" and "live": it checks the system against policies, security baselines, and deployment requirements before release. No readiness assessment, no deployment.[Module 7 sheet, p.432]
In plain words: the bank's AI governance policy is written to be technology-agnostic — it covers any AI system, whatever the law, industry, or tech — so when a new model arrives, the same readiness checklist applies instead of writing a new policy each time.

1.3 Understand deployment requirements

Choose the deployment environment

EnvironmentWhat it isPlain-words example
Cloud-basedModel runs on cloud infrastructure (AWS, Azure, GCP); scalable, pay-per-useA customer chatbot runs on the cloud so it can scale to 100k concurrent users during sales season
On-premiseModel runs on the organisation's own servers; full data controlA hospital runs its diagnostic AI on-premise so patient data never leaves the building
EdgeModel runs on local devices — phones, sensors, cameras; low latency, offlineA factory defect-inspection model runs on the camera itself, flagging defects in milliseconds without network access
GenAI — fine-tuned, RAG?Generative AI choices: fine-tune a foundation model, or use Retrieval-Augmented Generation to ground answers in your dataA legal assistant uses RAG so answers cite the firm's actual documents instead of hallucinating

Agentic architecture

Infrastructure

The platform, orchestration, and tool access that agents run on.

Example: an agent that books travel needs APIs to flights, hotels, and payments — each a new attack surface.

Risk models

Agents act autonomously — risk models must cover each action the agent can take.

Example: a refund agent's risk model limits it to refunds under $500; anything larger escalates to a human.

Governance frameworks — the 3 guardrail tiers

The sheet's tiered guardrails for agentic systems.

Tier 1: foundational guardrails · Tier 2: risk-based guardrails · Tier 3: societal guardrails (below)
TierGuardrail typeWhat it coversPlain-words example
Tier 1Foundational guardrailsBasic safety: the agent cannot exceed its permissions, access forbidden data, or act outside its purposeThe travel agent cannot book anything without a confirmed budget approval — a hard boundary baked into the system
Tier 2Risk-based guardrailsScoped by risk level of the action: higher-risk actions need more oversightLow-risk actions (answering FAQs) run free; high-risk actions (changing a delivery address) require human confirmation
Tier 3Societal guardrailsBroader societal impact: bias, manipulation, public harm, fundamental rightsA recruitment agent is reviewed for demographic fairness and cannot use protected attributes in decisions
Exam memory hook — the 3 guardrail tiersTier 1 = foundational (can't overstep), Tier 2 = risk-based (oversight scales with risk), Tier 3 = societal (impact on people and society). Agentic AI questions almost always test one of these three.[Module 7 sheet, p.432]

Other considerations

Applicable laws and policies

Map the legal landscape before deployment — PDPA, GDPR, EU AI Act, sectoral rules.

Example: a credit model must satisfy MAS fair-dealing guidelines plus PDPA consent rules.

System options

Consider alternatives — a different model, a simpler system, or no AI at all.

Example: a rules-based check may handle 90% of the use case with zero model risk.

Document appropriate uses of AI

Write down what the system may and may not be used for.

Example: the model card states the triage tool is for screening, not diagnosis.

Risk tolerance

The organisation's appetite for different harm types.

Example: the bank accepts credit-decision risk but not safety risk — tolerance is documented.

Risk assessment

Formal assessment of the deployment's risks, per the framework in place.

Example: a probability × severity matrix rates each failure mode of the chatbot.

Test, evaluate, verification, validation

The model must pass testing before release — verification (built right) and validation (right system built).

Example: the fraud model is verified against specs and validated on live-shadow data before go-live.

Package the model

Version, containerise, and document the model for deployment.

Example: the model ships as a versioned container with its model card, config, and dependencies pinned.

Make the model accessible for real-world use

APIs, interfaces, and integration points that users actually touch.

Example: the chatbot's API is documented, rate-limited, and monitored from day one.

Continuous monitoring

Monitoring starts at deployment, not after problems appear.

Example: drift alerts, error rates, and escalation paths are armed before the first real user arrives.
2
Curriculum block 2 · Lesson 1 (L1)

Own Proprietary Model vs Third-Party

The sheet's trade-off analysis: risks and opportunities of deploying your own model, managing third-party risks, and evaluating vendor/open-source agreement terms.

2.1 Risks — deploying your own proprietary AI model

Nature of data used

You own the data risk — quality, sensitivity, and lawful basis are your responsibility.

Example: if training data contains special-category data, the company carries the full GDPR exposure.

Transparency issues

Complex models may be hard to explain to regulators and users.

Example: a deep neural net's credit decisions resist the simple explanations the regulator asks for.

Data sources

Provenance risk: where the data came from, and whether it was lawfully obtained.

Example: a model trained on scraped data carries legal risk even if the model performs well.

Ownership challenges

Who owns the model, its training data, and its outputs — internally and with partners.

Example: an agency builds a model for a client — but who owns it when the contract ends?

Risk mitigation for AI use

Mitigations fall on you: no vendor to share the burden.

Example: bias findings must be fixed by your team — there's no vendor patch to wait for.

Increased requirements for proprietary models

Higher governance burden: documentation, testing, monitoring, and audit trails.

Example: the regulator expects full technical documentation — the in-house team must produce it all.

Additional liabilities — provider and deployer

You may carry both hats: liable as provider (building it) and deployer (using it).

Example: under the EU AI Act, the in-house team is both provider and deployer of the credit model — double obligations.

2.2 Opportunities — deploying your own proprietary model

Able to source data used to train the model

Full visibility of training data — no black-box provenance.

Example: the team can answer "what data trained this model?" from the lineage log in one screen.

Better fulfil governance reporting and regulatory requirements

You have the documentation and control to satisfy regulators directly.

Example: the AIA, model card, and risk register are all produced in-house, on demand.

Less susceptibility to security issues from open-source and third-party models

No inherited supply-chain or backdoor risk from code you didn't write.

Example: a known open-source model had a backdoor — the in-house model isn't affected.

Better opportunity for purpose fit

Train exactly for your use case, data, and domain.

Example: the bank's fraud model is trained on its own transaction patterns — off-the-shelf models underperform on them.
Exam memory hook — own model: 4 opportunitiesSource (know your data), Report (meet regulatory requirements), Secure (fewer third-party security issues), Fit (purpose-fit). The four opportunities all come from one thing: control.[Module 7 sheet, p.432]

2.3 Manage third-party risks

Context: third-party AI integrates into business operations — internal or external, including internal employee use of external tools.

RiskWhat it meansPlain-words example
Data lineage and traceabilityYou may not know where the vendor's training data came fromA vendor chatbot trained on scraped personal data — the customer's data flows into an unknown pipeline
Downstream issuesVendor changes, model updates, or shutdowns ripple into your systemsThe vendor retires a model version and the bank's chatbot suddenly behaves differently
Model output ownership and control — depending on licensingWho owns the outputs, and what the licence lets you do with themThe licence forbids using outputs to train a competitor model — the team must check before fine-tuning
Data handling and security risks — depend on development environmentYour data may be processed in environments you don't controlEmployees paste customer data into a public AI tool — it's now in the vendor's training pipeline
Quality and model performance may not fit the riskThe vendor's performance claims may not match your risk profileA vendor claims 98% accuracy, but on the bank's demographic mix it drops to 71% — unfit for the risk
In plain words — the context line: "integration into business ops — internal or external, internal employee use" — third-party AI risk applies even when only employees use the tool. An employee pasting client names into a public LLM is a third-party risk event.

2.4 Evaluate key terms & risks in vendor and open-source agreements

Term to evaluateWhy it mattersPlain-words example
Data considerationsHow the vendor uses your data — training rights, retention, sub-processingThe contract must state the vendor cannot train on your customer data
Security / safetyVendor's security posture, breach obligations, safety certificationsThe vendor must notify the bank within 24 hours of any breach affecting its data
Potential for biasDoes the vendor disclose bias testing and subgroup performance?The vendor's model card must include disaggregated performance by demographic group
Type of productSaaS, API, on-prem license, open-source — each has different risk and controlAn open-source model gives full control but zero support; a SaaS gives support but less control
Technical specsLatency, throughput, integrations, and infrastructure requirementsThe vendor's API must handle the bank's 5,000 requests-per-minute peak
Model performance resultsEvidence of performance — benchmarks, test results, limitationsThe vendor's benchmark scores are checked against the bank's own test data
Terms of usePermitted uses, restrictions, liability caps, termination rightsThe licence restricts use to internal purposes — deploying it for customers breaches the terms
Exam memory hook — the 7 agreement termsData, Security, Bias, Product type, Technical specs, Performance, Terms — "DSB-PTPT". If a scenario asks what to check before signing an AI vendor contract, these seven.[Module 7 sheet, p.432]
3
Curriculum block 3 · Lesson 2 (L2)

Release, Monitor & Maintain

Lesson 2 governs what happens after release: periodic assessment, performance monitoring, incident response when it fails, and the downstream-consequence guidelines.

3.1 Periodic assessment of the AI model

Performance

Continual checks of accuracy and quality against benchmarks (detailed next slide in the course).

Example: monthly scorecard of the fraud model's precision and recall against go-live baselines.

Reliability

The system is available and consistent when called upon.

Example: uptime above 99.5% and identical output for identical input, verified quarterly.

Safety

Four tool types: audits, red teaming, threat modeling, security testing.

Example: an annual red-team exercise probes the chatbot for jailbreaks and data leaks.
Exam anchor — the four safety toolsSafety assessment = audits (independent review), red teaming (adversarial probing), threat modeling (structured attack analysis), security testing (technical testing). All four appear in the sheet — expect a matching question.[Module 7 sheet, p.432]

3.2 Manage and monitor AI systems — assessing performance

Are there deviations in accuracy?

Accuracy drifting from baseline signals data or concept drift.

Example: fraud-model accuracy fell from 93% to 87% over a quarter — flagged for review.

Is the model making irregular decisions?

Odd outputs — distribution shifts, edge-case failures, or unexpected behaviour patterns.

Example: the pricing model suddenly quotes 10× normal for a product category — irregular decision.

Are there data drifts affecting performance?

Data drift, concept drift, and label drift — the three drift types from Module 6.

Example: after a marketing campaign, the customer mix shifts and the churn model's inputs no longer match training.

3.3 Risks and mitigations

MitigationWhat it meansPlain-words example
Use current best practicesFollow updated frameworks and standards, not last year's playbookThe team adopts the latest NIST AI RMF guidance in its monitoring playbook
Conduct red teaming exercisesAdversarial testing by a team that tries to break the systemA red team finds a prompt-injection path that makes the chatbot leak internal data — fixed before attackers find it
Consider bug bashing / bug bountiesPaid crowdsourced vulnerability discoveryThe company runs a bug bounty for its AI products; a researcher earns $5,000 for a jailbreak report
Document using model cards and standard documentationKeep model cards current as the system evolvesEvery model version ships with an updated model card; the audit trail shows what changed and when
Snapshot the algorithm and its outputsFreeze versions and output samples for forensics and comparisonBefore a retrain, the team snapshots the current model and 1,000 sample outputs for later comparison
Monitor risks from third partiesVendors and upstream models change under you — watch themThe vendor updates its foundation model; the team re-tests before allowing the new version into production

3.4 Changes to the model — the three drifts

DriftDefinitionPlain-words example
Model driftOverall degradation of model performance over timeThe fraud model's F1 falls from 0.91 to 0.78 — the model as a whole has drifted
Concept driftThe relationship between inputs and outputs changes"Free" used to mean spam; now a legitimate campaign uses it — the rule itself moved
Data driftThe distribution of input data changesCard-present fraud was the norm; now 80% of fraud is online — the input mix shifted
Exam memory hook — the three driftsModel drift = performance decays. Concept drift = the rule changes. Data drift = the inputs change. Same taxonomy as Module 6 — the sheet repeats it here under "changes to the model".[Module 7 sheet, p.432]

3.5 What to do when the AI system is not performing as it should

Use your incident response plan

The plan exists for this moment — activate it, don't improvise.

Example: the chatbot leaks a customer's data; the incident playbook starts the clock on containment and notification.

Identify the issue and understand to whom it needs to be reported

Know your escalation map: internal leadership, DPO, regulator, affected users.

Example: a PDPA notifiable breach must be reported to PDPC within 30 days — the team knows the route.

Understand risks based on third-party integration and third-party tools the AI integrates with

The failure may originate upstream or downstream in the toolchain.

Example: the pricing model went wrong because a third-party data feed changed format — the incident spans the vendor too.

Have the ability for a human to shut down the algorithm

The kill switch — a named human with a working stop mechanism.

Example: the operations lead has a single switch that takes the chatbot offline in 60 seconds, tested quarterly.

3.6 Assessing potential consequences & communicating updates

Resentment with poorly-implemented projects

Users distrust AI that was rolled out badly — reputational damage is real.

Example: a bank's buggy chatbot gave wrong answers for a week; customers now refuse to use it even after the fix.

False sense of safety and privacy

Users may over-trust the system's safety and share more than they should.

Example: a mental-health chatbot's disclaimer is buried — users believe it's confidential when it isn't fully.

Unintended consequences

Second-order effects nobody planned for.

Example: an anti-fraud model blocks legitimate elderly customers' transactions, causing distress and lost trust.

3.7 Guidelines — downstream consequences of AI research & development

GuidelineWhat it meansPlain-words example
Review potential downstream consequences early in R&DThink about harm before the model is finished, not afterA facial-analysis project is assessed for misuse potential at the research stage — and redesigned
Categorise AI research and downstream consequences by risk levelTier research by potential impactDual-use research is tagged high-risk; a benign recommender is tagged low-risk
Normalise discussions about downstream consequences of AI R&DMake harm-checking a routine part of every projectEvery project kickoff includes a 15-minute downstream-consequences discussion — standard agenda item
Be fully transparent and proactive in identifying negative downstream consequencesSurface harms early and publicly, don't bury themThe team publishes a known-limitations note when a model ships, including the biases they found
Develop common protocols for responsible product development, deployment, continuous improvementShared playbooks across the organisationA single responsible-AI playbook covers every product: develop, deploy, improve — with updates in clear language, freely available
Exam anchor — the protocols detailThe sheet adds two protocol specifics: information on updates to AI capabilities should be freely available and use clear language, and the protocols should address potential downstream consequences, including guidance, optimisation, and mitigation strategies.[Module 7 sheet, p.432]

3.7 Disclosures & transparency obligations

The sheet's disclosure block — when you must tell people the AI is there:

ObligationWhat it meansPlain-words example
Under most laws, must disclose any AI being usedTransparency about AI use is a baseline across most data-protection and AI regimesThe bank's chatbot must tell customers they are talking to an AI, not a human
EU AI Act — provider and deployer requirementsProviders owe Art. 50 transparency duties (chatbots, deepfakes, emotion recognition); deployers of Annex III high-risk AI that make decisions about natural persons must inform those persons they are subject to the system (Art. 26(11))A hospital's triage AI must inform patients their care decisions are assisted by the system
Some use cases or contexts require disclosureEven where the general law doesn't mandate it, the context may — sectoral rules, professional ethics, or contractual termsA recruitment agency discloses AI screening to candidates because the industry code requires it
Disclosure may be required so users can exercise rights or legal protectionsPeople need to know AI is involved to exercise appeal, contest, or data-protection rightsAn applicant can only appeal an AI-driven rejection if they know it was AI-driven — disclosure unlocks the right
In plain words: disclosure is the key that unlocks every other right. If users don't know an AI made the decision, they can't appeal it, challenge it, or ask for a human review — which is why "must disclose any AI being used" is the sheet's first line.

3.8 Incidents, issues and risks

In plain words: when the pricing model starts quoting nonsense, the team doesn't just check the model — the data team checks the feed, the vendor team checks the upstream API, legal checks notification duties, and the support team gathers user reports. The cause is a shared investigation.

3.9 Attributes of AI that can cause failures

The sheet lists six attributes of AI systems that cause failures in deployment — the same six failure modes from testing, now framed as ongoing deployment risks:

AttributeHow it causes failurePlain-words example
BrittlenessSmall input changes break the modelA sticker on a stop sign flips the vision model's answer
UncertaintyOverconfidence when the model should be unsure99% confidence on an ambiguous medical case
HallucinationsConfident fabrication by generative modelsA chatbot invents a refund policy that doesn't exist
Embedded biasTraining data skew surfaces in live decisionsA screening tool penalises group membership learned from history
Catastrophic forgettingFine-tuning destroys previously learned knowledgeAfter legal fine-tuning, the model forgets the maths it knew
False positivesFlags things that aren't real3× more legitimate transactions blocked as fraud
Exam memory hook — "BUH-ECF"The same six from Module 6's testing types appear here as failure attributes — brittleness, uncertainty, hallucinations, embedded bias, catastrophic forgetting, false positives. Know them in both framings: what you test for, and what fails in production.[Module 7 sheet, p.432]

3.10 Awareness of AI auditing & accountability issues

Assessments and audits are common mechanisms for accountability

AIAs, DPIAs, system audits, and model-card reviews are how organisations demonstrate they are answerable for their AI.

Example: the annual AI audit covers the AIA, bias tests, and compliance verification — the paper trail is the accountability.

Consider data protection rules for automated decision-making

GDPR Art. 22: the right not to be subject to solely automated decisions with legal or similarly significant effects — with exceptions (contract, law, consent) and required safeguards (human intervention, right to contest).

Example: an AI-only loan rejection triggers Art. 22 rights — the applicant must be able to obtain human review and contest the decision.

Automation in AI governance helps organisations stay competitive and meet regulations

Governance tooling — automated inventories, drift monitoring, compliance dashboards — scales oversight and keeps pace with regulation.

Example: an automated AI-inventory tool updates risk scores nightly, so the compliance team sees drift before the regulator asks.
Exam anchor — Art. 22 in an AI contextAutomated decision-making rules (GDPR Art. 22; Singapore PDPA's comparable provisions on decisions that produce legal or similar effects) apply regardless of whether the decision-maker calls it AI. The exam tests that human review must be meaningful, not a rubber stamp.[GDPR Art. 22; Module 7 sheet, p.432]
4
Interactive · 12 high-yield terms · tap to flip

Flip Cards — Deployment Terms

Tap a card to flip it. Term on the front, definition + exam tip on the back. If your browser doesn't support 3D flips, the cards still work — they cross-fade instead.

5
Self-test · Curriculum-based exam traps

Eight Exam Traps

TRUE / FALSE — every question drawn directly from the Module 7 sheet. Instant feedback, best score persists.

6
Exam-day readiness

Before Thursday — The Checklist

Tap each item as you master it. The goal: fill every box from memory, not recognition.

§
Citation ledger

Sources