Introduction
Hong Kong is becoming an important node for many mainland enterprises to arrange overseas servers, international data centres, regional operating headquarters and AI business pilots. Unlike the EU, which conducts risk-tiered regulation through a unified AI-specific law, and the mainland, which regulates in parallel through data compliance, algorithm filing, and generative AI and deep synthesis rules, Hong Kong has not yet formed a unified AI-specific law. Instead, it takes the Personal Data (Privacy) Ordinance (Cap. 486, hereinafter "PDPO") as its legal foundation, and the Office of the Privacy Commissioner for Personal Data, Hong Kong (hereinafter "PCPD" or the "Privacy Commissioner's Office") has gradually formed an AI governance rule system through model frameworks, compliance checks, employee usage checklists, AI Agent risk alerts and Deepfake toolkits.[1]
The "Artificial Intelligence: Model Personal Data Protection Framework" (hereinafter the "2024 Model Framework") released by the PCPD in 2024 has been in operation for nearly two years. This article focuses on the current state of governance in Hong Kong after the framework came into operation, and analyses the PCPD's regulatory functions, the legal nature of the 2024 Model Framework, the hardening path of soft-law rules, and the compliance priorities for Chinese enterprises deploying AI systems in Hong Kong. This article argues that the core of Hong Kong's AI governance is not "no regulation" or "light regulation", but an intermediate governance model centred on personal data protection, institutional accountability, risk assessment, human oversight and continuous auditing.[2]
01. Raising the Question: Hong Kong's AI Governance Has Entered the Post-Framework Review Stage
For enterprises using Hong Kong as a hub for going global, AI governance issues usually do not emerge at the time of company registration, but gradually surface in the course of server site selection, data centre deployment, international client operations, cross-border team collaboration and third-party AI tool procurement. Enterprises often first understand Hong Kong as an overseas operating platform, and then treat AI systems as technical tools to improve customer service, sales management, R&D collaboration or internal efficiency. But once an AI system processes customer data, employee data, user profiles, transaction behaviour, communication records or enterprise knowledge bases, Hong Kong's personal data protection rules will come into play.
The institutional feature of Hong Kong's AI governance is that it takes personal data protection law as its foundation, uses governance frameworks and scenario toolkits issued by the regulator as its operational path, and drives rule implementation through compliance checks and continuous auditing. The PCPD released the 2024 Model Framework in 2024 with the aim of assisting organisations in complying with PDPO requirements when procuring, implementing and using AI systems, and continuing the data governance values and ethical principles in its 2021 Guidance on the Ethical Development and Use of AI.[3]
As of 2026, the 2024 Model Framework is no longer merely a "newly issued guidance". From the 2025 employee usage checklist for generative AI and the 2026 Agentic AI risk alert to the 2026 compliance checks on AI use by 60 organisations, Hong Kong's AI governance has shown an evolution from principles to framework, from framework to scenarios, from scenarios to checks, and from institutional texts to audit evidence.[4]
Therefore, this article does not aim to "introduce the 2024 Model Framework", but takes the current state of governance in Hong Kong after its operation as its object of study, and discusses three questions: first, the legal status and boundaries of power of the PCPD in Hong Kong's personal data protection, AI governance and data compliance system; second, why the 2024 Model Framework, although not a statute, may still become a regulatory reference and compliance proof standard in enterprise AI deployment; third, how Chinese enterprises, when using Hong Kong as a hub for going global, should translate Hong Kong's AI governance requirements into internal systems, supplier management and continuous audit arrangements.
02. The Institutional Foundation of Hong Kong's AI Governance: AI Governance under Personal Data Protection Law
At the present stage, Hong Kong has not yet formed a unified AI-specific law similar to the EU's Artificial Intelligence Act (hereinafter the "AI Act"). Its AI governance does not take AI system admission licensing, algorithm filing or administrative approval as its main path, but rather, within the existing legal framework, forms compliance requirements for the use of personal data by AI systems through the PDPO, the six data protection principles, PCPD guidance and industry practice.[5]
The core of the PDPO is to regulate data users' collection, holding, processing and use of personal data. Its Schedule 1 establishes six Data Protection Principles (hereinafter "DPPs"), covering the purpose and manner of collection, accuracy and retention period, restriction on use, security measures, information disclosure, and data access and correction. Once an AI system involves personal data, whether for training, fine-tuning, retrieval augmentation, customer interaction, employee management or automated analysis, it needs to return to the framework of lawfulness, reasonableness, necessity, security, transparency and correctability required by the DPPs.[6]
From the perspective of governance structure, Hong Kong's AI compliance is highly bound up with data compliance. Data input, model processing, output use, permission access, log retention, supplier invocation and cross-border access in AI systems may all simultaneously trigger personal data protection, data security, information security, contractual liability and industry regulatory requirements. For cross-border operating enterprises, although the Hong Kong entity may only be the location of servers, the location of data centres or a regional operating centre, its role in system permissions, business purposes, customer notification, supplier contracts and data access arrangements is sufficient to make it bear personal data processing responsibilities.
In addition, it is not the case that Hong Kong has no cross-border data governance. Although section 33 of the PDPO has long not been fully implemented, the PCPD continues to issue recommendations and model contract tools relating to cross-border transfers; on the issue of cross-border flow of personal information in the Guangdong-Hong Kong-Macao Greater Bay Area, the PCPD has also emphasised the significance of the "Standard Contract for Cross-boundary Flow of Personal Information within the Guangdong-Hong Kong-Macao Greater Bay Area (Mainland, Hong Kong)" in rule alignment. This shows that Hong Kong adopts a governance path for cross-border data centred on contracts, transparency, safeguards and accountability, rather than simply laissez-faire.
03. The Legal Status and Regulatory Functions of the PCPD and Its Relationship with Other Authorities
An accurate understanding of the PCPD is the prerequisite for understanding Hong Kong's AI governance. The PCPD is not an industry association, nor an institution that merely provides publicity and educational materials. It is the statutory regulator for personal data protection in Hong Kong established and operated under the PDPO, with functions including supervising the implementation of the Ordinance, handling complaints, conducting investigations and inspections, issuing enforcement notices, publishing guidance and practical tools, conducting public education, and providing legal assistance to data subjects under specified conditions.[7]
Unlike the mainland's multi-department administrative regulatory system involving the Cyberspace Administration, the Ministry of Industry and Information Technology, the public security authorities and the market regulation authorities, the PCPD's positioning is closer to a data protection authority in the GDPR context. But Hong Kong's system has its own characteristics: the PCPD usually does not use prior approval or filing as its main tool, but promotes compliance through complaints, investigations, compliance checks, enforcement notices, guidance publication and case reports.
In the event of large-scale data breaches, privacy security incidents or infringement of personal data rights, there is a division of labour among the PCPD, the Hong Kong courts, the police and industry regulators. The PCPD is usually the core entry point for personal data compliance issues, responsible for complaint handling, investigation, inspection, enforcement notices and publication of investigation reports; the Hong Kong courts are responsible for civil compensation, judicial review and the trial of related criminal cases; the police and other law enforcement agencies handle criminal offences such as hacker attacks, fraud, identity theft, serious cybercrime or doxxing; and industry regulators in finance, healthcare, telecommunications and other sectors may impose additional regulation based on licensed business, consumer protection or industry security requirements.[8]
This distinction is particularly important for Chinese enterprises. Enterprises cannot simply assume that "if there is no AI-specific law, there is no AI regulation", nor can they understand PCPD documents as general publicity materials. Under Hong Kong's current system, whenever an AI system involves personal data, the PCPD is the most direct, most active and most rule-productive regulatory entry point.
04. The Nature of the 2024 Model Framework: Not a Statute, but Becoming a Regulatory Reference Standard
The legal nature of the 2024 Model Framework needs to be defined with care. It is not a statute passed by the Hong Kong Legislative Council, nor an administrative regulation that directly sets penalties; more accurately, it is an AI governance operational framework and regulatory expectations document issued by the PCPD based on its regulatory duties under the PDPO. But "not a statute" does not mean "no compliance significance". Under Hong Kong's institutional environment, in which principled legal obligations and regulatory guidance operate together, the framework has at least four practical functions.[9]
First, it translates the abstract obligations under the PDPO into operational requirements across the AI lifecycle. The PDPO requires data users to comply with principles such as lawful collection, purpose limitation, data security, transparency and data subject rights; the 2024 Model Framework further requires organisations to implement these principles in AI strategy, procurement, risk assessment, human oversight, model customisation, continuous monitoring and stakeholder communication.
Second, it is becoming the reference standard for the PCPD's AI compliance checks. On 19 May 2026, the PCPD published the results of compliance checks on the impact of AI use on personal data privacy by 60 organisations. The checks not only focused on whether the organisations complied with the PDPO when using AI systems to collect, use and process personal data, but also examined whether they had implemented the 2024 Model Framework and the recommendations of the employee usage checklist for generative AI, and assessed their overall AI governance performance.[10]
Third, it can become the evidentiary standard for enterprises to prove that they have exercised reasonable care. When an AI-related personal data incident occurs, whether the enterprise has conducted a personal data impact assessment, whether it has set up human oversight, whether it has carried out supplier reviews, whether it has retained logs and approval records, and whether it has trained employees will all affect its ability to defend itself in complaints, investigations, civil claims or customer due diligence.
Fourth, it may affect regulatory rectification and the assessment of legal liability. Under Hong Kong's personal data protection regime, a breach of the DPPs does not usually in itself automatically constitute a criminal offence; however, if a data user fails to comply with an enforcement notice issued by the PCPD, criminal liability may be triggered. Therefore, an enterprise's failure to establish an AI governance system in accordance with the Model Framework will not automatically constitute an independent violation of the law, but in the event of personal data leakage, unlawful use, processing beyond authority or harm to rights, it will increase the risk of being found to have failed to govern properly, failed to take reasonably practicable measures, or being required to accept rectification. [11]
05. Governance evolution after the operation of the 2024 Model Framework: from document publication to auditable compliance
When the 2024 Model Framework was published, its main contribution was to move AI governance from abstract ethical principles down to corporate governance processes. The Framework adopts a risk-based approach, provides concrete recommendations for Hong Kong enterprises intending to procure, implement and use AI systems, and covers the different situations of enterprises procuring off-the-shelf AI solutions and customising AI solutions according to their own needs. [12]
After nearly two years of operation, Hong Kong's AI governance has shown clear phased changes. First, from principles to a framework. In 2021 the PCPD issued the Guidance on the Ethical Development and Use of AI, emphasising the value of data governance and ethical principles; the 2024 Model Framework translates these principles into internal corporate governance structures, risk assessments, human oversight and continuous monitoring. Second, from framework to scenarios. The PCPD has continued to issue checklists, toolkits and risk alerts on employees' use of generative AI, AI chatbots, AI Agents, Deepfake and child protection, moving AI governance from a general framework into business scenarios. [13]
Third, from recommendations to inspections. In 2026 the PCPD's third round of AI compliance inspections covered 60 organisations, in industries including banking and finance, beauty, education, government departments, insurance, healthcare, utilities, retail, social welfare, telecommunications and transport, and extended to fields such as accounting, catering, innovation and technology, logistics and property management. The inspection results show that 57 organisations (95%) use AI in their daily operations; of these, 45 have used AI for more than one year, and 29 use three or more AI systems. This set of data shows that AI is already regarded by Hong Kong regulators as an ordinary compliance risk in daily operations, rather than a frontier technology issue for a minority of technology companies. [14]
Fourth, from institutional texts to audit evidence. In the future, enterprises will not only need to state "we have an AI policy", but also need to prove that the AI policy has been implemented, recorded, reviewed, trained on and audited. AI systems are characterised by continuous iteration: model versions, prompts, knowledge bases, plug-ins, APIs, supplier terms and permission configurations may all change the risk status. Therefore, AI governance should be designed as a continuous management mechanism, rather than a one-off pre-launch review.
06. The four major governance requirements under the 2024 Model Framework
(1) AI strategy and governance structure
The first part of the 2024 Model Framework is AI strategy and governance structure. Its core is not to require enterprises to formulate a macro technology vision, but to require enterprises to clarify the purposes of AI use, governance responsibilities, procurement mechanisms, approval processes, training arrangements and accountability structures. For enterprises using Hong Kong as a hub for going global, they especially need to answer: which businesses may use AI, and which businesses should be prohibited or restricted; who has the authority to approve the launch of new AI systems; how data roles are divided among the Hong Kong entity, the mainland parent company, overseas teams and suppliers; and who is responsible when AI output is erroneous or leaks secrets.
Enterprises should establish a cross-departmental governance structure that brings management, legal, compliance, information security, IT, business departments and procurement departments into a unified process. For AI projects involving personal data, customer rights and interests, employee evaluation or automated business decisions, pre-launch review, risk classification, approval records and periodic review mechanisms should be put in place.
(2) Risk assessment and human oversight
Risk assessment and human oversight are the key to Hong Kong's AI governance moving from "principled compliance" to "operable compliance". Enterprises need to identify whether AI systems collect or use personal data, whether they involve sensitive or high-impact data, and whether they affect individuals' rights and interests, customer opportunities, employment evaluation, transaction judgments or public services. For high-risk scenarios, human intervention, human review and human override mechanisms should be put in place, and AI output must not be used directly as the final decision.
In the Hong Kong context, risk assessment should not be limited to technical security testing, but should also cover issues such as the lawful source of personal data, changes in the purpose of use, data retention, third-party access, output bias, transparent notification, data subject access and correction, and incident response. If an enterprise cannot explain what data the AI system processes, whom it affects, who supervises it and how errors are corrected, its compliance risk will rise significantly.
(3) Model customisation, implementation and continuous monitoring
The third part focuses on model customisation, implementation and continuous monitoring. A typical deployment of Chinese enterprises in Hong Kong is not to develop a foundation model from scratch, but to procure third-party AI SaaS, call APIs, deploy privatised models, connect to enterprise knowledge bases, or connect internal documents with large models through Retrieval-Augmented Generation ("RAG"). In this case, the enterprise is both a user of the AI system and may bear higher governance responsibility due to customisation, configuration and business embedding.
Continuous monitoring includes at least data preparation, lawfulness of test data, spot checks of model output, assessment of version updates, permission control, log retention, supplier changes, handling of abnormal events, and exit and deletion mechanisms. In particular, when AI systems are connected to customer management systems, email, contract repositories, financial systems or employee databases, the permission matrix, data partitioning and the principle of least privilege should be treated as basic controls.
(4) Transparency, explanation and feedback mechanisms
The fourth part emphasises communication with stakeholders. AI governance cannot remain only in back-end security; it also requires front-end transparency, explainability, complainability and accountability. Enterprises should appropriately explain the circumstances of AI use, processing purposes, data categories, recipients and data subject rights in personal information collection statements, privacy policies, employee policies, customer agreements or business interfaces.
In scenarios involving automated analysis, customer segmentation, customer service replies, credit or risk judgments, HR screening, etc., enterprises should provide reasonable explanations and feedback channels. Explanation does not necessarily require disclosure of algorithm source code, but at a minimum it should enable individuals to understand for what purposes the AI system processes their personal data, how it affects their rights and interests, and how to make inquiries, corrections or complaints.
07. Three key AI deployment scenarios with Hong Kong as a hub for going global
(1) Deploying AI systems on Hong Kong servers, in data centres or at international operations centres
The first high-frequency scenario is where an enterprise places servers, data centres or international operations centres in Hong Kong, and uses a Hong Kong entity to undertake overseas customer service, regional marketing, order management, risk control analysis, knowledge base retrieval or internal operations support. In this case, the compliance issue is not whether the servers are located in Hong Kong itself, but where the data comes from, who controls it, who has permission to access it, whether personal data is involved, whether it is transmitted back to the mainland or a third place, and whether AI output affects individuals' rights and interests.
A more typical scenario that is also more prone to misjudgment is: the group establishes an entity in Hong Kong, and the Hong Kong entity holds or leases data centres and servers; overseas branches collect local customer, employee or partner data in their daily business and transmit the data to Hong Kong servers; the mainland parent company has access to, configuration rights over or substantive control over the data on the Hong Kong servers for group management, product R&D, risk control analysis or AI operations purposes; at the same time, mainland third-party cloud service providers, AI model suppliers or software technology service providers provide model invocation, algorithm optimisation, system maintenance or security operations services for the system. Under such a structure, enterprises cannot simply conclude responsibility allocation on the basis that "the data is in Hong Kong" or "the Hong Kong company holds the servers"; instead, they should return to the actual state of control by each subject over the purposes, methods, permissions and attribution of interests in the processing of personal data.
From the perspective of Hong Kong's PDPO, the first step is to identify who is the data user. If the Hong Kong entity determines the business use of the AI system, the data processing purposes, the data retention methods and the external service processes, it should generally be regarded as the core subject bearing personal data protection obligations under Hong Kong law. If the mainland parent company, although not directly holding the servers, substantively determines the system architecture, data analysis purposes, model selection, access permissions and processing rules, it may constitute a joint controller or substantive controller at the group governance level; even though the conceptual system of Hong Kong's PDPO is not entirely equivalent to the "controller/processor" dichotomy under the GDPR, enterprises still need to clarify the boundary of responsibility between the Hong Kong entity and the mainland parent company through internal agreements and data processing arrangements.
From the perspective of the laws of the places where overseas branches are located, the transmission of local personal data by overseas branches to Hong Kong servers may simultaneously trigger requirements under local data protection laws regarding cross-border data transfer, notification and consent, lawful basis, data processing agreements and security safeguards. Taking an EU member state branch as an example, if it transmits EU personal data to Hong Kong, it still needs to determine, in accordance with GDPR rules, obligations such as the cross-border transfer mechanism, the controller-processor relationship, data processing agreements and DPIA; taking Southeast Asian or other regional branches as examples, local personal data protection regulations and regulatory practice also need to be examined separately.
From the perspective of mainland law, if the mainland parent company remotely accesses, downloads, analyses or retrieves personal information on Hong Kong servers, or if mainland third-party service providers provide models, cloud services or technical operations for the Hong Kong system and actually come into contact with the relevant data, it is necessary to further determine whether this constitutes providing personal information to within China, processing activities initiated from within China, or data export, personal information processing, entrusted processing and security management obligations involving mainland personal information and important data. In particular, when the mainland parent company transmits mainland customer or employee personal information into the Hong Kong AI system, or transmits data to overseas model service providers through the Hong Kong system, the application of the mainland Personal Information Protection Law, Data Security Law and Cybersecurity Law cannot be automatically excluded on the ground that "the system is deployed in Hong Kong". [15]
Therefore, in this type of scenario, enterprises should at least complete five tasks: first, draw a data flow map across subjects, jurisdictions and systems, distinguishing data sources, data categories, access paths and output uses; second, form an intra-group data role matrix, clarifying the data processing links that the Hong Kong entity, the mainland parent company, overseas branches and third-party service providers each decide or execute; third, fix in contractual form obligations regarding entrusted processing, joint control, technical services, confidentiality, security incident notification, data deletion and audit cooperation; fourth, carry out a personal data impact assessment or similar risk assessment for high-risk AI processing, and retain approval, testing, human oversight and log records; fifth, separately examine the laws of Hong Kong, the mainland, the places where overseas branches are located and the places where third-party service providers are located, so as to avoid covering global data flows with conclusions drawn from a single jurisdiction.
For Chinese enterprises, placing AI systems in Hong Kong does not mean that compliance risks are cut off; on the contrary, the Hong Kong entity may become a concentrated node for personal data processing, cross-border access and international business responsibility in relation to AI systems. The real compliance judgment is not the physical location of the servers, but who determines the purpose of data processing, who controls the data, who configures system permissions, who bears the risk, and whether the enterprise can provide cross-jurisdictional, auditable governance evidence.
(II) Hong Kong entities procuring third-party AI SaaS, API or model services
The second type of scenario is a Hong Kong entity procuring or calling third-party AI SaaS, API, cloud model services, plug-ins or automation tools. Enterprises often understand this as ordinary IT procurement, but under the personal data protection framework, the admission of third-party AI tools is itself a compliance control point.
Enterprises should focus on reviewing whether the supplier uses inputs, outputs or logs to train models; where data is stored, backed up and accessed; whether there are sub-processors such as cloud service providers, model service providers, labelling service providers or plug-in service providers; how data and logs are deleted after termination of the service; whether encryption, isolation, access control, vulnerability response and incident notification are provided; and whether the enterprise enjoys rights to audit, make inquiries and obtain security certifications.
Employees' use of public generative AI tools can also be classified as a third-party AI tool admission issue. The PCPD's checklist and article on employees' use of generative AI emphasise that organisations should formulate clear internal policies or guidelines specifying the permitted scope of use, personal data protection, lawful and ethical use, prevention of bias, data security and the consequences of policy violations. Therefore, enterprises should not rely solely on employees' individual judgment to control AI inputs, but should conduct organisational management through tool whitelists, prohibited-input lists, approval processes and training records.[16]
(III) Deploying AI Agents or automated business agents in Hong Kong
The third type of scenario is an enterprise deploying AI Agents or automated business agents in Hong Kong. In March 2026, the PCPD issued a risk alert regarding OpenClaw and Agentic AI, pointing out that compared with AI chatbots usually used for text replies, summaries or content generation, Agentic AI generally has broader functions and may be deployed on local devices or servers with relatively high access permissions, reading and writing local files, allocating system resources, processing external services, and may even execute multi-step tasks on behalf of users according to preset workflows.[17]
There is also a very realistic commercial background for Chinese enterprises deploying AI Agents in Hong Kong: some overseas large-model services may not be directly and stably usable by users in mainland China due to geographical, account, payment, API policy or compliance restrictions. Enterprises therefore choose to establish entities, deploy servers, configure Agent gateways or automated calling layers in Hong Kong in order to access overseas large models, offshore SaaS services and international business systems. Such arrangements cannot be understood merely as technical routing or network access issues, but should be identified as AI system deployment and data processing activities organised, controlled or participated in by the Hong Kong entity. As long as the AI Agent processes personal data, corporate customer data or cross-border business data in the Hong Kong environment, it must still comply with the Hong Kong PDPO and the PCPD's regulatory requirements regarding AI systems, Agentic AI and personal data security.
The risk of AI Agents is no longer merely "incorrect output content", but extends to the level of system permissions, automated execution and cross-system data flows. The main risks highlighted by the PCPD include: relatively high default access permissions, which may access files, emails, device account credentials and content saved in browsers; vulnerabilities in system design or security controls that may create overall data security risks; and plug-ins or skills that have not undergone strict security review that may be implanted with malicious code and lead to account takeover or system compromise. When deploying AI Agents, enterprises should set minimum permissions, tiered authorisation, isolation of sensitive systems, manual confirmation at key nodes, plug-in whitelists, interface review, access logs, output records and regular risk reviews.[18]
Compared with mainland China, Hong Kong currently does not have a unified procedure for AI Agent deployment similar to generative AI service filing, algorithm filing, deep synthesis filing or dedicated ethical review. In other words, enterprises deploying AI Agents in Hong Kong generally do not need to make prior filings with or obtain approval from the PCPD merely because of the act of "deploying an Agent". However, this does not mean that there are no compliance obligations. Hong Kong's regulatory focus is: whether the Agent processes personal data, whether it complies with the data protection principles under the PDPO, whether reasonable and practicable security measures have been taken, whether there are deficiencies in transparency, permission control, supplier management and human oversight, and whether the enterprise can provide auditable evidence after a complaint, inspection or security incident occurs.
From the perspective of mainland law, if an AI Agent provides generative AI services to the public in mainland China, or affects mainland users through algorithm recommendation, deep synthesis or generative AI functions, it may still be necessary to make a judgment in light of mainland rules on algorithm recommendation, deep synthesis, generative AI services, security assessment, content security and labelling. In particular, where a mainland entity provides services to mainland users through a Hong Kong Agent, or where a Hong Kong Agent substantively serves mainland business scenarios, "deployment in Hong Kong" cannot automatically circumvent mainland regulatory requirements.[19]
From the perspective of future trends, Hong Kong may not replicate mainland China's filing system or the EU AI Act's market access model in the short term, but the PCPD has already brought AI Agents into its regulatory vision through the Agentic AI risk alert. As the permissions of AI Agents in enterprise systems deepen, Hong Kong rules are likely to continue to develop along the path of "soft-law toolkits, scenario-based checklists, compliance inspections and security alerts", imposing more detailed governance requirements on high-permission Agents, Agents connected to personal data systems, and Agents facing consumers or employees. Therefore, even if enterprises currently face no prior filing, they should establish in advance AI Agent inventories, permission matrices, tool-calling whitelists, manual confirmation nodes and continuous audit mechanisms.
08. Differences among the AI governance rules of Hong Kong, mainland China and the EU
Hong Kong AI governance should be understood between the mainland Chinese and EU regimes. Hong Kong is not an AI compliance vacuum, nor is it EU-style strong market access regulation, and it is even less similar to the mainland regulatory model characterised by filing, security assessment, content security and administrative supervision. It is more like an intermediate governance model centred on personal data protection, organisational accountability, regulatory guidance, compliance inspections and post-event investigations.[20]
The above differences mean that Chinese enterprises cannot directly transplant mainland compliance documents to Hong Kong, nor can they completely replace Hong Kong analysis with the high-risk system logic of the EU AI Act. Hong Kong places greater emphasis on accountability governance, transparent communication, risk assessment, human oversight, supplier management and continuous auditing in the process of personal data processing.
09. Conclusion: The key to Hong Kong AI governance is "auditable continuous compliance"
Following the release of the Model Framework in 2024, the refinement of rules on employees' use of generative AI in 2025, the Agentic AI risk alert and AI compliance inspections in 2026, Hong Kong AI governance has moved from principle advocacy into an auditable stage. Its core is not to require enterprises to stop using AI, nor to restrict innovation through high-intensity administrative access, but to require enterprises to be able to explain and prove: why the AI system is used, what data it processes, which subjects it affects, who approves it, who supervises it, how errors are corrected, how records are kept and how audits are conducted.
For enterprises using Hong Kong as a hub for going global, AI governance should produce at least six types of compliance deliverables: first, AI governance policies and division of responsibilities; second, an AI system inventory and data flow map; third, personal data risk assessment and human oversight mechanisms; fourth, third-party AI supplier review and contract terms; fifth, a system for employees' use of generative AI and admission of AI tools; sixth, records of logs, training, approvals, incident response and continuous auditing.
The real value of Hong Kong AI governance lies in its ability to serve as an institutional interface for Chinese enterprises connecting mainland compliance with international governance. If enterprises can establish an AI governance system in Hong Kong with reference to the PCPD framework, it will not only help reduce local personal data protection risks in Hong Kong, but also help enhance their compliance credibility when facing overseas customers, investors, suppliers, regulators and cross-border partners.
Therefore, Hong Kong is not an AI compliance vacuum, but an AI governance node where rule density, audit requirements and regulatory practice are taking shape. In the future, when Chinese enterprises use Hong Kong as a hub for going global, they need to move further from "where the servers are placed" to "how AI systems are governed, recorded and proven compliant".
Endnotes and reference documents
Questions and answers
Is the 2024 Model Framework legally binding?
The 2024 Model Framework is not a statute or regulation with direct penalties. However, it serves as a regulatory reference standard for PCPD compliance checks and can be used as evidence of reasonable care in investigations or disputes. Failure to follow it does not automatically constitute a violation, but may increase liability risk.
What are the compliance priorities for Chinese enterprises deploying AI in Hong Kong?
Enterprises should establish AI governance policies, conduct risk assessments, implement human oversight, manage third-party suppliers, control employee use of generative AI, and maintain continuous auditing records. They must also map cross-border data flows and address obligations under Hong Kong, mainland and other applicable laws.
Do AI Agents deployed in Hong Kong require prior regulatory approval?
No. Hong Kong does not have a filing or approval system for AI Agent deployment. However, if an AI Agent processes personal data, it must comply with the PDPO and PCPD guidance. The PCPD has issued risk alerts on Agentic AI, emphasizing permission controls, human oversight and auditable records.
How does Hong Kong's AI governance differ from mainland China and the EU?
Hong Kong relies on personal data protection law and soft-law guidance, without a unified AI law or prior filing system. The EU has a risk-tiered AI Act, while mainland China uses data compliance, algorithm filing and generative AI rules. Hong Kong's model is intermediate, focusing on accountability, risk assessment and continuous auditing.
Laws and cases cited
Laws
- Personal Data (Privacy) Ordinance (Cap. 486)
- Artificial Intelligence Act
- Personal Information Protection Law
- Data Security Law
- Cybersecurity Law
Policy documents
- Artificial Intelligence: Model Personal Data Protection Framework
- Guidance on the Ethical Development and Use of AI
- Standard Contract for Cross-boundary Flow of Personal Information within the Guangdong-Hong Kong-Macao Greater Bay Area (Mainland, Hong Kong)
References
- Personal Data (Privacy) Ordinance (Cap. 486), Chapter 486 of the Laws of Hong Kong; the official website of the Office of the Privacy Commissioner for Personal Data, Hong Kong, on the PDPO and the six data protection principles; Office of the Privacy Commissioner for Personal Data, Hong Kong, Artificial Intelligence: Model Personal Data Protection Framework, June 2024; Office of the Privacy Commissioner for Personal Data, Hong Kong, Checklist on Guidelines for the Use of Generative AI by Employees, 2025; Office of the Privacy Commissioner for Personal Data, Hong Kong, The PCPD Issues Alert over the Privacy Risks of OpenClaw and Agentic AI and Reminds Organisations and the Public to Use AI Safely, 16 March 2026; Office of the Privacy Commissioner for Personal Data, Hong Kong, Abuse of AI Deepfakes: Toolkit for Schools and Parents, 2025.
- Office of the Privacy Commissioner for Personal Data, Hong Kong, Artificial Intelligence: Model Personal Data Protection Framework, June 2024; this framework centres on personal data protection governance when organisations procure, implement and use AI systems, and sets out recommendations on AI strategy and governance, risk assessment and human oversight, model customisation and system management, and stakeholder communication.
- Office of the Privacy Commissioner for Personal Data, Hong Kong, Artificial Intelligence: Model Personal Data Protection Framework, June 2024; Office of the Privacy Commissioner for Personal Data, Hong Kong, Guidance on the Ethical Development and Use of Artificial Intelligence, 2021.
- Office of the Privacy Commissioner for Personal Data, Hong Kong, Checklist on Guidelines for the Use of Generative AI by Employees, 2025; Office of the Privacy Commissioner for Personal Data, Hong Kong, The PCPD Issues Alert over the Privacy Risks of OpenClaw and Agentic AI and Reminds Organisations and the Public to Use AI Safely, 16 March 2026; Office of the Privacy Commissioner for Personal Data, Hong Kong, The Privacy Commissioner’s Office has Completed Compliance Checks on 60 Organisations Regarding the Impact of the Use of Artificial Intelligence on Personal Data Privacy, 19 May 2026.
- Regulation (EU) 2024/1689, Artificial Intelligence Act; Personal Data (Privacy) Ordinance (Cap. 486), Chapter 486 of the Laws of Hong Kong; the official explanation of the Office of the Privacy Commissioner for Personal Data, Hong Kong on the PDPO and the six data protection principles.
- Schedule 1 to the Personal Data (Privacy) Ordinance (Cap. 486) sets out six data protection principles (Data Protection Principles), covering the purpose and manner of collection, accuracy and retention period, use, security, openness, and access and correction.
- The provisions of the Personal Data (Privacy) Ordinance (Cap. 486) on the functions and powers of the Privacy Commissioner, including complaints, investigations, inspections, enforcement notices and legal assistance; the official explanation of the Office of the Privacy Commissioner for Personal Data, Hong Kong on complaints, investigations, enforcement notices and legal assistance.
- Sections 38, 50, 66 and 66B of the Personal Data (Privacy) Ordinance (Cap. 486); the official explanation of the Office of the Privacy Commissioner for Personal Data, Hong Kong on complaint handling, investigations, enforcement notices, legal assistance and enforcement reports; where hacking, fraud, identity theft, doxxing or licensed industry business is involved, the Police, the Hong Kong courts and the relevant industry regulators in finance, healthcare, telecommunications and other sectors may still intervene separately.
- Office of the Privacy Commissioner for Personal Data, Hong Kong, Artificial Intelligence: Model Personal Data Protection Framework, June 2024; the six data protection principles in Schedule 1 to the Personal Data (Privacy) Ordinance (Cap. 486).
- Office of the Privacy Commissioner for Personal Data, Hong Kong, The Privacy Commissioner’s Office has Completed Compliance Checks on 60 Organisations Regarding the Impact of the Use of Artificial Intelligence on Personal Data Privacy, 19 May 2026; Office of the Privacy Commissioner for Personal Data, Hong Kong, Checklist on Guidelines for the Use of Generative AI by Employees, 2025.
- Office of the Privacy Commissioner for Personal Data, Hong Kong, Data Protection Principles in the Personal Data (Privacy) Ordinance — From the Privacy Commissioner’s Perspective; sections 50 and 64A of the Personal Data (Privacy) Ordinance (Cap. 486).
- Office of the Privacy Commissioner for Personal Data, Hong Kong, Artificial Intelligence: Model Personal Data Protection Framework, June 2024.
- Office of the Privacy Commissioner for Personal Data, Hong Kong, Guidance on the Ethical Development and Use of Artificial Intelligence, 2021; Office of the Privacy Commissioner for Personal Data, Hong Kong, Artificial Intelligence: Model Personal Data Protection Framework, June 2024; Office of the Privacy Commissioner for Personal Data, Hong Kong, Checklist on Guidelines for the Use of Generative AI by Employees, 2025; official materials of the Office of the Privacy Commissioner for Personal Data, Hong Kong on the safe use of AI chatbots; Office of the Privacy Commissioner for Personal Data, Hong Kong, The PCPD Issues Alert over the Privacy Risks of OpenClaw and Agentic AI and Reminds Organisations and the Public to Use AI Safely, 16 March 2026; Office of the Privacy Commissioner for Personal Data, Hong Kong, Abuse of AI Deepfakes: Toolkit for Schools and Parents, 2025.
- Office of the Privacy Commissioner for Personal Data, Hong Kong, The Privacy Commissioner’s Office has Completed Compliance Checks on 60 Organisations Regarding the Impact of the Use of Artificial Intelligence on Personal Data Privacy, 19 May 2026.
- Personal Information Protection Law of the People’s Republic of China, Data Security Law of the People’s Republic of China, Cybersecurity Law of the People’s Republic of China; Measures for the Standard Contract for Outbound Transfer of Personal Information, Provisions on Promoting and Regulating Cross-Border Data Flows and other mainland rules relating to personal information processing and cross-border data.
- Office of the Privacy Commissioner for Personal Data, Hong Kong, Checklist on Guidelines for the Use of Generative AI by Employees, 2025; official press releases, articles and promotional materials of the Office of the Privacy Commissioner for Personal Data, Hong Kong on employees’ use of generative artificial intelligence.
- Office of the Privacy Commissioner for Personal Data, Hong Kong, The PCPD Issues Alert over the Privacy Risks of OpenClaw and Agentic AI and Reminds Organisations and the Public to Use AI Safely, 16 March 2026.
- Office of the Privacy Commissioner for Personal Data, Hong Kong, The PCPD Issues Alert over the Privacy Risks of OpenClaw and Agentic AI and Reminds Organisations and the Public to Use AI Safely, 16 March 2026; this document sets out safe-use tips regarding the risks that Agentic AI may access local files, emails, account credentials, content saved in browsers, external services and plugins.
- Provisions on the Administration of Algorithm Recommendations for Internet Information Services, Provisions on the Administration of Deep Synthesis of Internet Information Services, Interim Measures for the Administration of Generative Artificial Intelligence Services, Basic Safety Requirements for Generative Artificial Intelligence Services; where providing services to the public within China, algorithm recommendations, deep synthesis, generative AI services, content security, labelling and security assessments are involved, a separate assessment should be made in light of the mainland rules.
- Hong Kong: the Personal Data (Privacy) Ordinance (Cap. 486), the Office of the Privacy Commissioner for Personal Data's Artificial Intelligence: Model Personal Data Protection Framework and its AI-related checklists, toolkits and compliance check materials; Mainland China: the Personal Information Protection Law of the People's Republic of China, the Data Security Law of the People's Republic of China, the Cybersecurity Law of the People's Republic of China, the Provisions on the Administration of Algorithm Recommendations for Internet Information Services, the Provisions on the Administration of Deep Synthesis of Internet Information Services, and the Interim Measures for the Administration of Generative Artificial Intelligence Services; the European Union: Regulation (EU) 2016/679, General Data Protection Regulation; Regulation (EU) 2024/1689, Artificial Intelligence Act.
