MAIE5103 Chap.10 AI Governance Frameworks and Release Decisions
AI Governance Frameworks and Release Decisions
Four regimes, four kinds of authority
The lectures compare four instruments, and the first rule is to cite each with its jurisdiction and date.
The NIST AI Risk Management Framework 1.0, of 26 January 2023, is voluntary, with a generative AI profile added in July 2024. The EU AI Act entered into force on 1 August 2024 as binding regulation organised by risk tier, with many duties applying from 2 August 2026. Mainland China's interim measures for generative AI took effect on 15 August 2023 and cover filing, security, content and labelling for public services.
Hong Kong, as of the lecture, has no dedicated AI Act and works through the Personal Data (Privacy) Ordinance and the privacy commissioner's guidance.
Govern, map, measure, manage
The NIST framework organises risk work into four functions. Govern sets owners, authority, risk tolerance, resources and accountability, and cuts across the rest.
Map covers purpose and non-uses, the people affected and the ways the system could fail. Measure covers valid targets and tests, group and system evidence and uncertainty. Manage covers mitigation, monitoring, response and the decision to pause, roll back or retire. The lecture stresses that these are iterative functions, not a checklist or a certification.
Hong Kong's 2024 model framework has four parts of its own and recommends levels of human involvement matched to risk.
Documents and the release decision
Datasheets, model cards and decision logs matter when they change a review, a decision or a handover.
They feed a release decision with three outcomes: go for a bounded use with monitoring, a limited pilot to resolve a stated uncertainty, or no-go where the purpose cannot be justified or a harm cannot be controlled. Running a pilot never suspends privacy or accountability duties, or the need for a safe fallback.
What this chapter covers
- 01
Pairing every rule with its jurisdiction and date
- 02
The NIST AI Risk Management Framework and its generative profile
- 03
The EU AI Act's risk tiers and staged duties
- 04
Mainland China's interim measures for generative AI
- 05
Hong Kong's PDPO and the privacy commissioner's model framework
- 06
Govern, map, measure and manage
- 07
Datasheets, model cards and decision logs as interfaces
- 08
Go, limited pilot or no-go
Worked example · free
Correct three governance claims in a draft report
- 2Correct the claim about NIST.
- 2Correct the claim about the EU AI Act.
- 2Correct the claim about Hong Kong.
Key terms
- AI Risk Management Framework
- NIST's voluntary 2023 framework that organises AI risk work into the Govern, Map, Measure and Manage functions.
- Risk Tier
- A category in the EU AI Act that determines which duties apply to a system, from prohibited practices to high-risk obligations.
- Limited Pilot
- A release decision that allows bounded use in order to resolve a stated uncertainty, while privacy duties, accountability and a safe fallback all still apply.
- Human-in-Command
- The level of human involvement Hong Kong's model framework suggests where neither human-in-the-loop nor human-out-of-the-loop fits.
- Decision Log
- A record of why a release was approved, which trade-offs were accepted, who owns monitoring and what triggered any rollback.
- Non-use
- A use a system is explicitly not intended for, mapped before release so that misuse can be recognised and refused.
AI Governance Frameworks and Release Decisions FAQ
What are the four functions of the NIST AI Risk Management Framework?
Govern, Map, Measure and Manage. Govern covers owners, authority, risk tolerance, resources and accountability and runs across the others. Map covers purpose, non-uses, stakeholders and failure paths. Measure covers valid targets, tests, evidence and uncertainty. Manage covers mitigation, monitoring, response and the decision to pause, roll back or retire.
The framework is voluntary and the functions are iterative, not a checklist.
Does Hong Kong have an AI law?
Not a dedicated one, as of the lecture. AI systems that handle personal data fall under the Personal Data (Privacy) Ordinance, and the privacy commissioner has issued ethical AI guidance in 2021, a model personal data protection framework in 2024, and agentic AI privacy guidance noted for August 2026. The framework's recommendations are guidance rather than legislation.
How do I decide whether an AI system is ready to release?
Work through the lecture's release questions: is the purpose legitimate and was a non-AI alternative compared, do the targets and tests fit the actual use, are residual harms, appeals and owners explicit, and are monitoring and rollback funded and tested? Then choose go for a bounded use, a limited pilot to resolve a named uncertainty, or no-go. The lecture presents this as a teaching tool, not an official decision tree.
Assessment move
Build a four-row table of the regimes with jurisdiction, date, binding or voluntary, and one thing each asks of a deployer, and test yourself on it until the dates come without effort. Then practise the release decision on one system per week, writing the three outcomes and which you chose. Governance answers lose marks mainly through wrong dates and 'must' attached to guidance, and both are fixed by the table.
Working through AI Governance Frameworks and Release Decisions in MAIE5103? Sia is AskSia’s AI Arts and Humanities tutor — ask any MAIE5103 AI Governance Frameworks and Release Decisions question and get a clear, step-by-step explanation grounded in how MAIE5103 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.