Every AI Policy Ever · open in the interactive register

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

Artificial Intelligence Risk Taxonomy and Assessment Framework

人工智慧風險分類框架

Record TW-02

Issued by
Ministry of Digital Affairs (MODA)
Jurisdiction
Taiwan
Level · region
National · East Asia
Published / adopted
2026-07-07
Language
Chinese (Traditional)

Summary

Framework made by MODA under Article 16 of the AI Basic Act; sector regulators must follow its workflow when writing risk-based AI rules. It lists 20 risk subtypes in three groups: technical design flaws, such as dangerous capabilities; post-deployment operation and human-AI interaction, such as deepfake fraud, cyberattacks and unauthorized actions by AI agents; and societal-structural and environmental impacts, such as cognitive warfare. Its four steps are use case inventory, risk identification, assessment and response. Use cases that may seriously harm national security, fundamental rights, life and safety, property, social order or the environment are high-risk.

Language

English translation by Ministry of Digital Affairs (MODA) Open it. The Chinese (Traditional) text is authoritative.

Safety topics

risk-based classificationgeneral-purpose / frontier AIdeepfakes & synthetic medianational security & military

Status notes

Version 1.0 is dated 7 Jul 2026 (ROC 115-07-07); MODA's English version carries the same version and date. The framework applies primarily to civilian AI use cases and expressly excludes military uses, which follow national defence laws. Where a sector regulator assesses a use case as high-risk, it must require warnings (Art. 5(2) of the AI Basic Act) and define liability and remedies (Art. 17(1)); where a use case's risks under Art. 5(1) cannot be managed with existing technical means, the regulator must restrict or prohibit it under Art. 16(2). The Ministry will review the list of risk subtypes periodically against new AI risk research and international standards.

Sources

Open the official source https://moda.gov.tw/major-policies/ai/governance/19244

Secondary source: https://www-api.moda.gov.tw/File/Get/moda/zh-tw/UieEfux5Yv0iJuR

Local copy

Archived. Last attempt 2026-10-18 00:44:50 UTC

Official source (third-party copy, not published) archive/TW-02/2026-10-11/official.html
Retrieved 2026-10-11 14:03:01 UTC
SHA-256 d2f3e2b1f2648dbfee61282c0f6ce1966c2b39262f7b6ab8e67c020406ddb16c
Secondary source (third-party copy, not published) archive/TW-02/2026-10-11/secondary.pdf
Retrieved 2026-10-11 14:03:06 UTC
SHA-256 9b28b24e441534a9dd047609f6439d8e281d249f8e82ba845f2d11da52923d84

Verification

A
The instrument's own official text, or an official register, was stored and confirms the title, date and status. checked 2026-10-18 · official-url-fetched
Recheck of 18 Oct 2026. Stored and read: MODA's Chinese page on the framework (moda.gov.tw, copy of 11 Oct 2026, re-checked unchanged on 18 Oct 2026), which names MODA's Department of Digital Strategy (數位策略司) as publisher and links the framework PDF (page created 2026-03-23, updated 2026-07-08); the Chinese framework PDF, version 1.0 (www-api.moda.gov.tw, copy of 11 Oct 2026, unchanged on 18 Oct 2026), dated 中華民國115年7月7日 (7 Jul 2026); MODA's English page on the framework (moda.gov.tw/en, 18 Oct 2026; created 2026-04-30, updated 2026-09-17), titled 'AI Risk Taxonomy and Assessment Framework'; and the English PDF it links (18 Oct 2026), 'Artificial Intelligence Risk Taxonomy and Assessment Framework', Version 1.0, July 7, 2026, with the same structure, 20 risk subtypes, steps and high-risk test as the Chinese text. The English version has no statement on its standing; the framework was made in Chinese, so the English one is recorded as MODA's translation. The title now follows MODA's English text, and the summary is written from the Chinese text with the English version's terms. The Chinese PDF is stored as the record's secondary source, so it counts as another official source; the date is also on the cover of MODA's English version. The earlier notes' reference to MODA's AI Evaluation Center rests on no stored file and was removed, as was the remark that the page's creation date might point to an earlier draft.
Second check: Independent check by Claude Code (claude-sonnet-5-5), 2026-10-18: 9 field(s) supported.

History

  1. 2026-10-18english_version: – → {"kind": "government translation", "url": "https://www-api.moda.gov.tw/File/Get/ — the record had none; MODA publishes an English version of version 1.0, now stored2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)
  2. 2026-10-18notes: Version 1.0 is dated 7 Jul 2026 (ROC 115-07-07). The MODA page was created 23 Ma → Version 1.0 is dated 7 Jul 2026 (ROC 115-07-07); MODA's English version carries — the reference to MODA's AI Evaluation Center (AIEC) rests on no stored file and was removed; the hedge about an earlier draft was removed (page dates are in verification.note); the civilian scope and military exclusion (2.2), the measures for high-risk and unmanageable use cases (3.4.1(b) iii and iv), the periodic review (3.2.3) and the English version added2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)
  3. 2026-10-18summary: Framework issued under Article 16 of the AI Basic Act to guide sector regulators → Framework made by MODA under Article 16 of the AI Basic Act; sector regulators m — rewritten from the Chinese text with the terms of MODA's English version (risk subtypes, the three risk-type names, the four steps); it now states that regulators must follow the workflow (2.1)2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)
  4. 2026-10-18short_title: Taiwan AI Risk Classification Framework → Taiwan AI Risk Taxonomy and Assessment Framework — follows the new English title2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)
  5. 2026-10-18title: Artificial Intelligence Risk Classification Framework (v1.0) → Artificial Intelligence Risk Taxonomy and Assessment Framework — MODA's English version of the framework is titled 'Artificial Intelligence Risk Taxonomy and Assessment Framework' (its English page: 'AI Risk Taxonomy and Assessment Framework'); the previous title was a rendering of 人工智慧風險分類框架. The version label moved to notes2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)

Related records

Taiwan AI Basic Act Taiwan · In force

Citation

Ministry of Digital Affairs (MODA). Artificial Intelligence Risk Taxonomy and Assessment Framework. 2026-07-07. https://moda.gov.tw/major-policies/ai/governance/19244. Status: Published (non-binding, current); verified 2026-10-18 (level A). Local copy archive/TW-02/2026-10-11/official.html retrieved 2026-10-11 14:03:01 UTC, SHA-256 d2f3e2b1f2648dbfee61282c0f6ce1966c2b39262f7b6ab8e67c020406ddb16c. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record TW-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.