Ethical AI Framework (Hong Kong) v2.0
Tags: Frameworks, Regulatory
TL;DR
- Hong Kong’s Digital Policy Office released a comprehensive Ethical AI Framework (v2.0, Dec 2025) covering principles, governance, lifecycle, and assessment.
- It standardises how organisations plan, build, deploy, and monitor AI, with a rigorous AI Application Impact Assessment at 4 lifecycle checkpoints.
Why it matters for HK marketers: It gives a clear, government-backed playbook to deploy AI confidently in campaigns while meeting accountability, privacy, and safety expectations.
What this framework includes
- Ethical AI Principles: 12 principles, with 2 performance principles—Transparency & Explainability; Reliability, Robustness & Security—foundational to all others.
- AI Governance: A three-lines-of-defence model spanning project teams, PSC/PAT oversight, and IT Board/CIO review, optionally advised by an Ethical AI Committee.
- AI Lifecycle: 6 practical stages—Project Strategy, Project Planning, Project Ecosystem, Project Development, System Deployment, System Operation & Monitoring—aligned with SDLC.
- AI Practice Guide: 18 practice areas mapped to the lifecycle, from sourcing to model monitoring.
- AI Application Impact Assessment: A templated, question-based assessment with risk gating, lifecycle-linked questions, and stakeholder impact considerations.
Who should use it
- IT Planners & Executives: Embed ethical AI from strategy and portfolio planning through sourcing.
- System Analysts, Architects, Data Scientists: Apply concrete practices for data, model design, deployment, and ongoing operations.
Alignment and scope
- Applies beyond government—v2.0 revises government-specific terms for general organisational use.
- Complements existing project management and IT security standards.
12 Ethical AI Principles.
6 AI Lifecycle stages.
4 assessment review checkpoints across the lifecycle.
3 lines of defence in AI governance.
So what for marketers
Use this as your default governance stack for any AI-powered campaign: run the assessment at the 4 checkpoints, align vendor and data choices to the 12 principles, and document PSC/PAT approvals to protect brand trust.
---
Additional content from market data:
TL;DR
- Serves as a governance foundation for AI in marketing alongside applicable ordinances, regulator guidance, contracts, and platform policies.
- Guides privacy‑by‑design, vendor/data location assessments, and safeguards for fair treatment and IP/confidentiality.
Why it matters for HK marketers: It translates Hong Kong’s policy stance into concrete guardrails to embed at use‑case design, not as a last‑minute legal fix.
Role in Marketing Compliance
- Anchor use‑case screening for: personal data, direct marketing, cookies/identifiers, cross‑border data transfer, regulated products, vulnerable audiences, employment data, biometrics/likeness, copyrighted material, automated consumer interactions.
- Drive privacy‑by‑design: lawful purpose, data minimisation, accuracy, retention limits, access security, and appropriate notices/choices/consents.
- Vendor/data checks: where prompts/logs/outputs are stored; training/use for service improvement; ability to disable training; data recipients; deletion processes; subcontractors/overseas processing.
- IP/confidentiality: confirm rights for source uploads and commercial use of outputs; check restrictions on training/scraping/grounding/style imitation, reuse, output ownership, attribution, retention, indemnities.
- Fairness: assess whether segmentation/personalisation/generative content could discriminate, stigmatise, or create unjustified differences in price, opportunity, visibility, or treatment.
Embedded in Playbook Examples
- Loyalty program personalisation: minimise data, exclude sensitive free text, use enterprise‑approved environment with training disabled, document retention/deletion, check overseas processing, obtain Privacy/InfoSec approvals.
So what for marketers
Adopt the framework’s checklist at the brief stage and document approvals; campaigns crossing borders should also test against the strictest applicable market rule.
← Back to Knowledge Base