Every AI Policy Ever · open in the interactive register

Published (non-binding, current)Non-binding · Guidelines / framework

Hong Kong Generative Artificial Intelligence Technical and Application Guideline

Record HK-02

Issued by
Digital Policy Office, HKSAR Government (drafted with the Hong Kong Generative AI Research and Development Center)
Jurisdiction
Hong Kong SAR (China)
Level · region
Sub-national · East Asia
Country
Hong Kong SAR (China)
Published / adopted
2025-04-15
Last updated
2025-12
Language
English, Chinese

Summary

Non-binding guidance for generative AI developers, service providers and users on risks such as hallucination, bias, deepfakes, jailbreaking and data leakage. It sets a governance framework with five dimensions (privacy, IP, crime prevention, reliability, system security) and a four-tier risk model (unacceptable, high, limited, low). It assigns role-based duties, including content labelling and watermarking of high-risk outputs, independent audits, pre-deployment risk assessment and human oversight matched to impact.

Safety topics

general-purpose / frontier AIrisk-based classificationtransparency & labellingdeepfakes & synthetic mediahuman oversight

Status notes

Released by the Digital Policy Office on 15 Apr 2025; the current text is version 1.1 (December 2025). The DPO commissioned the Hong Kong Generative AI Research and Development Center (HKGAI), established in 2023 with AIR@InnoHK funding, to study and suggest codes and guidelines, and promulgated the guideline after considering its findings; it says it will update the guideline regularly. Where the Chinese and English versions differ, the English version prevails. It is published on the same DPO page as the Ethical AI Framework (HK-01).

Sources

Open the official source https://www.digitalpolicy.gov.hk/en/our_work/data_governance/policies_standards/ethical_ai_framework/

Secondary source: https://www.digitalpolicy.gov.hk/en/our_work/data_governance/policies_standards/ethical_ai_framework/doc/HK_Generative_AI_Technical_and_Application_Guideline_en.pdf

Local copy

Archived. Last attempt 2026-10-18 07:40:31 UTC

Official source (third-party copy, not published) archive/HK-02/2026-10-11/official.html
Retrieved 2026-10-11 14:04:38 UTC
SHA-256 f3fe4ff8caf13d528455dbf02009f6adfc21f85ed6852c1b2f330e8ca93ad2a0
Linked document (PDF) (third-party copy, not published) archive/HK-02/2026-10-11/document-1.pdf
Retrieved 2026-10-11 14:04:46 UTC
SHA-256 8cf14d321262aae177a739f8f54bd5d7373ffad58a004224654fd05d2adcacd2
Linked document (PDF) (third-party copy, not published) archive/HK-02/2026-10-11/document-2.pdf
Retrieved 2026-10-11 14:04:51 UTC
SHA-256 a0ec7fa02e162e73f7f10939896815425e49e80ef29cacf770bb8de370eecf0b
Secondary source (third-party copy, not published) archive/HK-02/2026-10-11/secondary.pdf
Retrieved 2026-10-11 14:04:53 UTC
SHA-256 8cf14d321262aae177a739f8f54bd5d7373ffad58a004224654fd05d2adcacd2

Verification

B
Another official source, such as an announcement by the issuer, confirms the title, date and status. checked 2026-10-18 · alternative-official-url-fetched
Refresh of 18 Oct 2026 under the rule that level A needs the instrument's own text or an official register. Re-read: the guideline's own text, version 1.1 dated December 2025 (digitalpolicy.gov.hk PDF, stored 11 Oct 2026, unchanged on 18 Oct 2026), which confirms the title, the DPO's commissioning of HKGAI, its purpose as codes and practices for reference by all sectors, the regular updates, the English-prevails clause and the content of the summary; and the DPO's page 'AI and Data Ethics' (stored 11 Oct 2026, unchanged on 18 Oct 2026), which says the DPO promulgated the guideline and links the PDF. Stored again in this run: the HKSAR Government press release of 15 Apr 2025 (info.gov.hk), which dates the first release to that day. Neither the guideline, whose cover gives only the date of version 1.1, nor the DPO page dates the first release, so date_published rests on the press release, an announcement, and the record is level B. The PDF has no version history, so what changed in version 1.1 is not stated.
Second check: Independent check by Claude Code (claude-sonnet-5-5), 2026-10-18: 6 field(s) supported.

History

  1. 2026-10-18verification.level: A → B — level computed from the evidence by tools/pipeline.py2026-10-18-refresh-level-rule-2 · Claude Code (claude-opus-5-5)
  2. 2026-10-18notes: Current text is version 1.1 (December 2025). The document does not give the rele → Released by the Digital Policy Office on 15 Apr 2025; the current text is versio — release date and HKGAI's founding and AIR@InnoHK funding added from the press release; the remarks on why date_published was empty and that the guideline was 'reportedly first issued in April 2025 (not verified)' concern the research and were removed; the English-prevails clause added from the PDF2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)
  3. 2026-10-18date_published: – → 2025-04-15 — was null; the HKSAR Government press release of 15 Apr 2025 says the DPO released the guideline that day2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)

Related records

Hong Kong Ethical AI Framework Hong Kong SAR (China) · Published (non-binding, current)

Citation

Digital Policy Office, HKSAR Government (drafted with the Hong Kong Generative AI Research and Development Center). Hong Kong Generative Artificial Intelligence Technical and Application Guideline. 2025-04-15. https://www.digitalpolicy.gov.hk/en/our_work/data_governance/policies_standards/ethical_ai_framework/. Status: Published (non-binding, current); verified 2026-10-18 (level B). Local copy archive/HK-02/2026-10-11/official.html retrieved 2026-10-11 14:04:38 UTC, SHA-256 f3fe4ff8caf13d528455dbf02009f6adfc21f85ed6852c1b2f330e8ca93ad2a0. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record HK-02.

Improve this record

Know of an amendment, a status change or an error? Report news or an error on GitHub; the form opens with this record filled in.