Japan's 2026 privacy law amendment: what changes for companies
In short
- The amending act passed on 10 July 2026 and was promulgated on 17 July 2026. It comes into force on a date to be fixed by cabinet order, within two years of promulgation.
- It relaxes consent requirements for some data uses while tightening the rules on higher-risk data and the enforcement against serious breaches. It is not simply a loosening or simply a tightening.
- The new exception for producing statistical information does not mean personal data may be used freely for AI. It is limited to uses that genuinely stop at general, non-individualised analysis.
- A surcharge is introduced, but not for every breach. It targets serious, large-scale violations with an economic motive.
- As at the time of writing, the cabinet orders, commission rules and guidelines are still being finalised. Settled statutory direction and still-to-be-specified detail need to be kept apart.
On 17 July 2026 Japan promulgated an act amending the Act on the Protection of Personal Information (APPI). It is the substantial revision produced by the statute's three-yearly review cycle.
For companies operating in Japan, the reach is wide: the consent rules for producing statistical information including some AI development, data about children under 16, facial feature data, the position of processors, breach notification, misuse of information that can be used to reach an individual, opt-out transfers, and a new surcharge.
The amendment is neither a deregulation for AI nor a simple tightening of privacy law. It makes certain data uses workable while raising the consequences for high-risk handling and for serious breaches.
Passed and promulgated — but the detail is still being written
The bill was submitted to the Diet on 7 April 2026, passed on 10 July and was promulgated on 17 July 2026.
Commencement is by cabinet order, within two years of promulgation. So not all of the amended provisions are in force yet.
The Personal Information Protection Commission is continuing to prepare the cabinet orders, commission rules and guidelines that fill in the operational detail. An outline of what will be set by subordinate legislation was published on 26 August 2026, and a series of "basic approach" papers followed in September and October.
For planning purposes, separate what the statute has settled from what subordinate rules will still specify.
| Theme | Direction | Effect on companies |
|---|---|---|
| Enabling appropriate data use | Consent rationalised for certain statistical production and for handling clearly not contrary to the individual's intention | New options for AI development, data collaboration and research — subject to purpose and governance conditions |
| Risk-based rules | Under-16s, facial feature data, processors and breach response revised | Service design, consent flows, processing agreements and privacy policies need review |
| Preventing misuse | Rules on information that enables contact with a specific individual even where it is not personal information; opt-out tightened | Compliance review for advertising data, identifiers and list trading |
| Effective enforcement | Stronger orders and penalties, cooperation requests to third parties, and a surcharge | Higher financial exposure; governance needed for large-scale data use |
The statistical production exception — relevant to AI, but narrow
The change that has drawn the most attention is the relaxation of consent requirements where data is used only to produce statistical information.
The amendment allows third-party provision of personal data, and acquisition of publicly available sensitive personal information, without the individual's consent in certain cases where it is secured that the data will be used only to produce statistical information. Government material states expressly that this covers AI development that can be characterised as statistical production.
The limit matters more than the permission. The exception turns on producing general, general-purpose analytical results that sever the correspondence to individuals. Where a third-party transfer is involved, there are also mechanisms binding provider and recipient to the stated purpose.
So it does not license identifying individuals from a trained model and marketing to them, scoring individuals, or repurposing the underlying personal data for ordinary service operation. Labelling a project "AI development" does not engage the exception.
Before relying on the exception for an AI project
- What exactly is being acquired or transferred, and which category does it fall into?
- Can the purpose genuinely be characterised as production of statistical information only?
- Is the system designed so that the model and its outputs are not used to act on specific individuals?
- Where there is a third-party transfer, how are the recipient's purpose and onward transfer constrained by contract?
- Which conditions remain to be specified by cabinet order, rules or guidelines before commencement?
Handling "clearly not contrary to the individual's intention"
New grounds for dispensing with consent are added for use beyond the stated purpose, acquisition of sensitive personal information, and third-party provision.
The concept is that consent is not required where, judged by the circumstances of acquisition, the handling is clearly not contrary to the individual's intention and clearly does not harm their rights or interests.
The example to have in mind is a transfer that is inherent in delivering the very service the individual asked for.
It is not a general convenience exception. "It is useful to us" and "the privacy policy is drafted widely" do not engage it. The circumstances of acquisition, the individual's reasonable expectation, the necessity of the handling and the effect on rights all have to be worked through.
Life, body and property; public health; research
The existing exceptions for protecting life, body or property and for public health are made easier to use. Under the current statute the condition that it be difficult to obtain consent causes practical problems; the amendment widens this to include cases where there is a reasonable ground for not obtaining consent.
It is also made explicit that institutions providing medical care fall within the academic research exception, which matters for medical and research data.
A wider exception does not reduce the security obligations. Access control, purpose management and transfer management scale with the sensitivity of the data regardless.
Children under 16
The amendment makes explicit that, where the individual is under 16, consent and certain notifications are directed to the legal representative.
It also relaxes the conditions for requesting cessation of use of retained personal data relating to a person under 16, and introduces a duty to give priority consideration to the best interests of the individual when handling minors' personal information.
For apps aimed at children, education services, games, social platforms, commerce and location services, this is not a privacy-policy edit. Age verification, the parental consent flow, default profile visibility, advertising and recommendation, and retention periods all sit at the service-design level.
Facial feature data
Specific rules are introduced for facial feature data used in facial recognition and identification.
The amendment requires certain matters about the handling of facial feature data to be made known, relaxes the conditions for requesting cessation of use, and prohibits third-party provision under the opt-out mechanism.
The reason is that a face cannot be changed like a password and supports continuous tracking, so the consequences of leakage or misuse are durable.
Any company using cameras in stores, access control, identity verification, events or marketing analytics should inventory what is displayed at the point of collection, the stated purpose, retention, third-party provision, processors, and how a cessation request would be handled.
Processors — "it is outsourced" is not an answer
The rules on entrusting data processing to cloud providers, analytics vendors, BPO operators and AI vendors are revised.
The amendment organises and clarifies the obligations of an entity that receives personal data for processing, with the central point being that it must not process beyond the scope entrusted to it. Where the method and the measures are specified in the contract and the entrusting party supervises, the allocation of obligations is set out.
This is where AI services sit. Whether a vendor is processing customer personal data to deliver the service, or also using it to improve its own general models, changes the legal characterisation.
So processing agreements and AI terms of service need to be explicit about the purposes for which input data may be used, whether it goes to sub-processors, retention, deletion, and whether it is used for training. The contractual mechanics are in the ten points to check in an AI contract.
Breach notification, calibrated to risk
Notification to affected individuals is adjusted: where there is little risk to their rights and interests, the duty to notify is relaxed.
This is not permission to treat incidents casually. It is a move toward balancing the cost of notification against the protection actually needed, based on what was exposed.
The practical consequence is that an incident response procedure which applies the same mechanical response above a record count is no longer sufficient. The process needs to be able to assess the type of information, whether it was encrypted, whether a third party could obtain it, and how it could be misused.
Information that is not personal data but can still reach a person
The amendment introduces rules prohibiting improper use and improper acquisition of information that is not personal information but can be used to contact or target a specific individual — by advertising display, messaging or otherwise.
Phone numbers, email addresses, addresses and cookie identifiers combined with other information fall into this territory.
The effect is that the formal question "is this personal information?" no longer settles whether data may be used freely. Advertising technology companies, data brokers and anyone trading lists or handling online identifiers should take this seriously.
Opt-out transfers
The opt-out mechanism — transferring personal data to third parties without consent, subject to stopping on request — is tightened.
The provider is placed under a duty to confirm the recipient's identity and purpose of use, and the recipient is required not to give false explanations.
The background is the use of purchased lists in fraud. For companies that sell or supply data, the question is how far confirmation goes beyond collecting a signed declaration.
Orders, penalties and third-party cooperation
Enforcement is strengthened. The conditions for issuing an order are revised so that correction can be required promptly rather than only after a graduated process, and the Commission may order measures necessary to protect rights and interests, including notifying affected individuals and publication.
A basis is also added for requesting a third party who assists a violation to take measures to stop it — for example, a service provider whose platform is used for unlawful handling.
On the criminal side, the scope and penalties for improper provision of personal information databases are revised, and a penalty is introduced for obtaining personal information by deception.
The surcharge is not a general revenue-based penalty
The surcharge has attracted the most attention, and the most misunderstanding.
It is not a scheme under which any APPI breach attracts a percentage of turnover. Government material frames it as targeting serious violations involving large volumes of personal data where there is an economic incentive, and where individuals' rights and interests have been harmed.
The covered conduct, the mental element and the scale threshold are to be specified in subordinate legislation. In an approach published by the Commission in October 2026, a figure of 1,000 affected individuals was indicated as one measure of the scale element for certain violations.
The amount is to be based on the financial benefit obtained through the violation.
The practical implication is not to manage everything to the same intensity. It is to keep decision-making and legal review records for the data uses that are large in scale and tied to revenue — monetised data use, third-party transfers, and acquisition of data from outside.
What changes for AI companies
The amendment cuts both ways for AI businesses. The statistical production route opens space for development; improper acquisition or use brings materially higher enforcement exposure.
So the useful posture is not "AI training is exempt" or "public data is free," but examining the data flow stage by stage.
Scraping public web data, for example, needs separate treatment of sensitive personal information, the source site's terms, whether the purpose is genuinely limited to statistical production, whether individuals are later acted upon, and the risk of re-identification from model outputs.
Fine-tuning and retrieval-augmented generation on customer data raise the processor question above: whether the AI vendor is processing on instruction or using the data for its own purposes.
Ten things worth doing before commencement
- Update the data map covering personal information, personal data and personal-related information
- For each AI and analytics use, identify the consent basis or the exception relied on
- Identify projects that might rely on the statistical production exception
- Check age verification and parental consent for services with users under 16
- Inventory systems that acquire or use facial feature data
- Confirm with processors, cloud and AI vendors what inputs are used for, including sub-processing and training
- If you make opt-out transfers, revisit how recipients are verified
- Build an incident process that can assess risk, not just count records
- Put approval and record-keeping around large-scale monetised use of personal data
- Re-review the privacy policy, consent screens, processing agreements and internal rules once the subordinate rules are settled
Not a policy rewrite at the last minute
The amendment touches acquisition, use, third-party provision, processing, incidents and enforcement.
Editing the privacy policy shortly before commencement will not be enough if the systems and data flows underneath have not been examined.
In an AI or data business in particular, which data, from whom, to whom, for what purpose, and what remains after training needs to be a shared description across engineering, business and legal — not three different ones.
My own view is that this amendment is better approached as an inventory of data flows and allocated responsibility than as a diff against the statutory text.
In short
The 2026 amendment enables certain data uses while tightening the rules on high-risk handling and serious breaches, at the same time.
A new route exists for AI development that can be characterised as statistical production. It is not a general exemption, and the purpose, the nature of the data, transfers, processing and effects on individuals all still have to be worked through.
Children's data, facial feature data, opt-out transfers and the surcharge can affect service design and board-level decisions directly.
Since the subordinate rules are still being finalised, the practical sequence is to inventory data flows now against the settled direction of the statute, and to put the detail into contracts, internal rules and interfaces once it is fixed.