Japan has long regulated personal data through guidance and administrative orders, with money penalties only at the end of a long escalation. That is about to change. The Diet passed amendments to the Act on the Protection of Personal Information (APPI) on 10 July 2026, and they were promulgated on 17 July 2026 as Act No. 56 of 2026, according to A&O Shearman's summary. The Personal Information Protection Commission (PPC) has since outlined how it will implement them.
A note on sourcing: the PPC's English pages we could reach confirm the reform process but not the final statutory text. The details below rest on law-firm analysis. Some outlets report different passage dates, and we use the 10 July and 17 July dates that A&O Shearman reports. Readers should check the consolidated text on the PPC site when it is published.
The strongest case for the new fines
The case for a surcharge is real. The PPC could previously only issue recommendations and orders, and criminal penalties applied only to firms that ignored an order. A company that scraped or traded personal data at scale could keep the proceeds of the violation and face no direct monetary consequence. When deterrence depends on a violator disobeying a regulator first, it is weak. Japan is also an outlier among large economies on this point: the GDPR, Korea's PIPA and others already let regulators fine directly.
What the amendments do
According to A&O Shearman, the PPC can order a surcharge equal to the money or other consideration a company received from the violating conduct. A Cabinet Order will set the calculation method. The criminal-penalty and improper-acquisition changes take effect on 17 January 2027, and the main reforms take effect on a date set by Cabinet Order, no later than 17 July 2028.
The design choice matters. A surcharge pegged to gain is a disgorgement model, not a revenue-percentage model like the GDPR's. It takes the profit out of violating the law without turning an ordinary compliance error into an existential threat. That is the more proportionate option and the one we favour. Reports that the surcharge will be aimed at serious violations that harm individuals' rights are consistent with this approach, but we could not verify the threshold against a primary PPC document.
The PPC has also reportedly asked for a 10.4% budget increase to prepare the implementing rules and guidelines. We have not confirmed that figure from a primary source. A regulator that gains a new fining power needs the staff to run fair procedures, and that is a reasonable use of money.
The AI-training carve-out
The provision most relevant to innovation is a new exception that allows providing personal information to third parties solely for "statistical creation." This includes producing analytical outputs such as AI training data. A&O Shearman says it applies only where the risk to individuals is low and the activity is specified by PPC rules. Recipients may use the data only within the disclosed scope, and onward disclosure is prohibited.
This is a sensible middle path. It does not declare AI training lawful in every case, and it does not leave developers to rely on legal uncertainty either. It lets consent-free uses that pose little risk proceed under defined conditions. The open question is the PPC's rulemaking. If the rules define "low risk" narrowly, or require documentation out of proportion to the risk, the exception will be a headline with little effect on real development. If the rules are clear and testable, Japan could offer AI developers a more predictable legal basis than much of Europe does.
Biometrics and children
The amendments create a category of "specified biometric personal information." This covers identifiers that are easy to obtain and whose collection may not be noticeable, such as faces captured by cameras. Operators must give advance notice of purposes (Article 21-2), and opt-out provision to third parties is prohibited (Article 27(2)). Individuals can request suspension of use even absent unlawful processing (Article 35(7)). Covert face capture is a genuine harm, and a notice-first rule is a proportionate response.
For children under 16, consent or notification generally goes through a legal representative (Article 40-2). Minors can exercise suspension and erasure rights without the usual preconditions (Article 35(9)), and firms must give primary consideration to the child's best interests (Article 58-3).
The risk here is the one civil-liberties groups have raised about child-safety measures elsewhere. The EFF argues that laws framed as protecting young people can hinder their ability to organise and speak online. A parental-consent default for every under-16 can push services to wall off teenagers entirely, or to collect more identity data to verify age. Both outcomes cut against the privacy goal. The PPC should state in its rules how firms can verify age with minimal data, and should avoid any expectation of blanket verification.
What to watch
- The Cabinet Order formula. Gain-based calculation must be defined clearly. Vague attribution of profit to a violation would invite large, contestable surcharges.
- Procedure. Notice, hearing and appeal rights should be set out before the first order issues.
- The statistical-use rules. These will determine whether the AI exception has practical value.
- Age assurance. How firms confirm a user is under 16 will decide whether the child provisions protect privacy or erode it.
Japan's approach is, so far, closer to proportionality than many of its peers. It avoids turnover-based fines and creates an AI pathway. The details left to the Cabinet Order and PPC rules will decide whether that holds, and the 2027 and 2028 effective dates leave time to get them right.