Every AI Policy Ever · open in the interactive register

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

Guide to Evaluation Perspectives on AI Safety in the Healthcare Sector: Toward the Realization of Trustworthy AI

ヘルスケア領域におけるAIセーフティ評価観点ガイド ～Trustworthy AI（信頼できるAI）の実現を目指して～

Record JP-11

Issued by
Japan AI Safety Institute (J-AISI)
Jurisdiction
Japan
Level · region
National · East Asia
Published / adopted
2026-04-03
Language
Japanese, English

Summary

Sector guide by the Japan AI Safety Institute's Healthcare Sub-Working Group that applies the ten evaluation perspectives of J-AISI's Guide to Evaluation Perspectives on AI Safety to healthcare. It covers AI providers whose products build on pre-trained text-generating LLMs and are not Software as a Medical Device, and for each perspective, from control of toxic output and prevention of misinformation to data quality and verifiability, sets out healthcare-specific risks, actual cases and example evaluation items, including the risk that such products are used as de facto medical devices. It explains evaluation in five phases from product design to deployment and operation, with a checklist.

Safety topics

evaluation & testingsafety institutegeneral-purpose / frontier AI

Scope

Borderline scope Written by the Healthcare Sub-Working Group, a group of companies within J-AISI's Business Demonstration Working Group, together with a working group of the Japan Digital Health Alliance; included because J-AISI convened the group and publishes the guide as one of its outputs.

Status notes

Written by the Healthcare Sub-Working Group of the Business Demonstration Working Group that J-AISI set up, led by Ubie, Inc. with other companies as members and Mitsubishi Research Institute as secretariat, in collaboration with the Working Group on Generative AI (SubWG-B) of the Japan Digital Health Alliance (JaDHA). J-AISI's list of outputs dates the release 2 April 2026; the Japanese and English covers are dated 3 April 2026. J-AISI also published summaries in both languages and, in Japanese, Markdown versions of chapters 3 and 4 for use with AI services and practical examples of AI safety evaluation with prompts and agent skills. Products that qualify as Software as a Medical Device are left to the approval and certification framework of the Pharmaceuticals and Medical Devices Act. The guide is a living document, to be updated as generative AI and regulation evolve.

Sources

Open the official source https://aisi.go.jp/output/output_information/260402/

Local copy

Archived. Last attempt 2026-10-18 00:45:55 UTC

Official source (third-party copy, not published) archive/_runs/2026-10-14-discover-leads-2026-10-14/C016/2026-10-14/official.html
Retrieved 2026-10-14 13:11:17 UTC
SHA-256 951778896a6fbaa2c4f7f426bf53d9ac4941d1fe5fc3e1c23bc2d3330ec3f2b8
Linked document (PDF) (third-party copy, not published) archive/_runs/2026-10-14-discover-leads-2026-10-14/C016/2026-10-14/document-1.pdf
Retrieved 2026-10-14 13:11:35 UTC
SHA-256 b86d121c089ebf57a36ccdabfa1e18f3f9d7728e7c21eda6b6484c42a430d5c7
Linked document (PDF) (third-party copy, not published) archive/_runs/2026-10-14-discover-leads-2026-10-14/C016/2026-10-14/document-2.pdf
Retrieved 2026-10-14 13:11:39 UTC
SHA-256 e9daae8da0d5f86bb330dd6a2e65f701447bb02e907ddd26a486c2e418894c5f

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: J-AISI's bilingual release page on aisi.go.jp; the Japanese guide version 1.0 (118 pages, cover 令和 8 年 4 月 3 日, with its development structure, the members of the sub-working group and the update history in chapter 7; copy of 18 Oct 2026); the English guide (98 pages, 'April 3, 2026') and the English summary (13 pages, 'April 3, 2026'); and J-AISI's list of Information outputs (copy of 18 Oct 2026), which dates the release 2026年4月2日. They confirm the title, the Japanese title with its subtitle, the issuer and authors, the scope, the dates and the content; the summary is checked against the Japanese text. The English edition carries no translation notice and is listed beside the Japanese edition, so it is treated as an authentic text. Version 1 gave the sub-working group as part of the issuing body and had no scope_note; following the rule for reports that an authority publishes and a group it convened wrote, the issuing body is now J-AISI, the scope_note names the group and the notes name the authors. The list of outputs shows no later version of the guide.
Second check: Independent check by Claude Code (claude-sonnet-5-5), 2026-10-18: 5 field(s) supported.

History

  1. 2026-10-18scope_note: – → Written by the Healthcare Sub-Working Group, a group of companies within J-AISI' — added under the same rule, naming the sub-working group and the Japan Digital Health Alliance's working group2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)
  2. 2026-10-18notes: J-AISI's list of outputs dates the release 2 April 2026; the Japanese and Englis → Written by the Healthcare Sub-Working Group of the Business Demonstration Workin — authorship added from chapter 7 (SWG led by Ubie with company members, secretariat Mitsubishi Research Institute, with JaDHA's SubWG-B); the release page lists the Markdown files of chapters 3 and 4 only in Japanese, so that sentence was corrected2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)
  3. 2026-10-18issuing_body: Japan AI Safety Institute (J-AISI), Healthcare Sub-Working Group of the Business → Japan AI Safety Institute (J-AISI) — the guide is a J-AISI output written by a sub-working group of companies that J-AISI set up (7.1, page 117 of the Japanese text); under the rule for reports an authority publishes that a group it convened wrote, the issuing body is the authority and the notes name the authors2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)
  4. 2026-10-18original_title: ヘルスケア領域におけるAIセーフティ評価観点ガイド → ヘルスケア領域におけるAIセーフティ評価観点ガイド ～Trustworthy AI（信頼できるAI）の実現を目指して～ — the Japanese cover carries the subtitle ～Trustworthy AI（信頼できる AI）の実現を目指して～, which the English title already renders ('Toward the Realization of Trustworthy AI'); added so that the two titles correspond2026-10-18-recheck-east-asia-1 · Claude Code (claude-opus-5-5)
  5. 2026-10-14Added to the register — found in pipeline run 2026-10-14-discover-leads-2026-10-14, candidate C016 (JP/ai-bodies)2026-10-14-discover-leads-2026-10-14 · Claude Code (claude-opus-5-5)

Related records

J-AISI Evaluation and Red Teaming Guides Japan · Published (non-binding, current)

Japan AISI Japan · Published (non-binding, current)

Citation

Japan AI Safety Institute (J-AISI). Guide to Evaluation Perspectives on AI Safety in the Healthcare Sector: Toward the Realization of Trustworthy AI. 2026-04-03. https://aisi.go.jp/output/output_information/260402/. Status: Published (non-binding, current); verified 2026-10-18 (level A). Local copy archive/_runs/2026-10-14-discover-leads-2026-10-14/C016/2026-10-14/official.html retrieved 2026-10-14 13:11:17 UTC, SHA-256 951778896a6fbaa2c4f7f426bf53d9ac4941d1fe5fc3e1c23bc2d3330ec3f2b8. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record JP-11.

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.