Japan’s PPC issues an Art. 147 APPI caution to OpenAI: keep special care-required data out of training corpora and publish the purpose of use in Japanese
Subject: OpenAI, L.L.C. and OpenAI OpCo, LLC
What happened
As generative AI services spread in Japan, the Personal Information Protection Commission published an alert on their use on 2 June 2023, and at the same time published the summary of a caution it had issued on 1 June 2023 under Art. 147 APPI to OpenAI, L.L.C. and OpenAI OpCo, LLC, the developer and provider of ChatGPT.
How it was handled
The caution has two limbs. First, without prior consent, do not acquire special care-required personal information about ChatGPT users or non-users (save for the statutory exceptions), and implement four measures for information collected for machine learning: take the necessary steps so that collected information does not contain special care-required personal information; as soon as possible after collection, take measures to minimise any such information that may be contained in it; where such information is nonetheless found, delete it or render the individual unidentifiable as immediately as possible and before processing the data into a training dataset; and where a data subject or the PPC requests or instructs that such information not be collected from a specific site or third party, comply unless there is a legitimate reason to refuse. In addition, where a user has opted out of having prompts used for machine learning, special care-required personal information in those prompts must not be processed without legitimate reason. Second, the purpose of use of personal information concerning both users and non-users must be notified or published in Japanese, to users and non-users alike.
Outcome
The caution also records that it reflects the concerns the Commission had clearly identified at the time, and that if new concerns emerge it may take additional action as needed.
What it means for Chinese exporters
- · Japan’s "special care-required personal information" (its sensitive-data category) is prohibited in principle at the point of acquisition — unlike Art. 9 GDPR, the regulated act is acquiring it. Companies building training corpora by crawling need filters on the collection side, not only on the output side, and must retain the filter rules and hit logs.
- · The regulator expressly requires the cleaning to happen before the data is processed into a training dataset. That fixes the engineering order: raw corpus landed → sensitive-data detection and deletion → training set built, with auditable boundaries and records between the steps; retrofitting afterwards is not accepted.
- · A data subject or the regulator can name a site and tell you to stop collecting from it. You need the ability to stop crawling by domain or source quickly and to trace and delete the corresponding data from existing corpora, or the "comply unless there is a legitimate reason" duty cannot be met.
- · The purpose of use must be published in Japanese and must cover non-users. Many Chinese products abroad publish only an English privacy policy and only notify registered users — in Japan neither is sufficient. Make "local language plus non-user coverage" the default requirement in every target market.
- · A caution is not a fine, but it is the record that precedes later administrative action. On receiving one, build an internal remediation ledger and completion evidence — it becomes the key defence if the same issue is raised again.
Sources (2)
- Japan PPC alert on the use of generative AI services, including the caution issued to OpenAIPersonal Information Protection Commission, Japan (PPC) · published 2023-06-02 · retrieved 2026-09-12
- Japan PPC published document: summary of the caution issued to OpenAI (Annex 2)Personal Information Protection Commission, Japan (PPC) · 别添 2 · published 2023-06-02 · retrieved 2026-09-12
Content as of 2026-09-12 · corpus 2026.09|Compiled from public official sources with links and review dates for compliance decision support. Rules change; have licensed professionals review before external disclosure, regulatory filing or contractual commitments.