Corporate LawLawyer-Reviewed
Q.An employee entered confidential company information into ChatGPT for work. What should the company do? Whether a report is required under Japan's APPI, the effect on trade secrets and partners, discipline and prevention
A.
First establish what was entered, into which service and under what settings. Personal data entered without consent and used for AI training may breach the APPI, and incidents within Article 7 of the Rules must be reported to the PPC and notified to individuals. Base any discipline on the work rules and proportionality, and set AI usage rules promptly.
Read more ▶
If you discover that an employee has been entering confidential company information into ChatGPT or another generative AI service, start by establishing what was entered, when, into which service and under what settings. Japan's Personal Information Protection Commission (PPC) has warned that entering prompts containing personal data without the individuals' consent may breach the Act on the Protection of Personal Information (APPI) where the data is handled for purposes other than producing the response, such as machine learning by the AI provider. If the incident amounts to a leak and falls within certain categories, such as one involving sensitive personal information, the company must report it to the PPC and notify the individuals (APPI, Art. 26). The company should also check the effect on its trade secrets and on confidentiality agreements with business partners, and decide any disciplinary action on the basis of the work rules and proportionality.
When this becomes an issue
- A salesperson pasted client names and deal details into an AI tool to draft proposals.
- A developer entered confidential source code into an AI tool to fix bugs.
- An HR officer used AI to summarise interview notes containing staff appraisals and medical history.
- An employee used a personal account to translate a draft contract received from a business partner.
1. Establish the facts first
The first step is not to decide on discipline in a hurry but to establish the facts. Confirm the following through interviews and records:
- What was entered (personal data, technical information, a partner's information) and how much
- When and how often it was entered, and which service was used
- Whether it was a corporate plan contracted by the company or the employee's personal account
- Whether the settings allowed the provider to use inputs for training, and whether the history can be deleted
- Whether the output was sent to anyone outside the company
How input is handled varies greatly with the type of service and the contract. Check the terms of use, privacy policy and administrator settings, and keep screenshots.
2. Where personal data was included
A business handling personal information must specify the purpose of use as far as possible (APPI, Art. 17(1)) and must not, without the individual's consent, handle personal information beyond what is necessary to achieve that purpose (Art. 18(1)). It must also take necessary and appropriate measures to keep personal data secure, including preventing leaks, loss or damage (Art. 23).
In its "Alert on the Use of Generative AI Services" of 2 June 2023, the PPC made the following points for businesses:
- Before entering a prompt containing personal information, confirm that doing so is within what is needed to achieve the specified purpose of use.
- If a business enters a prompt containing personal data without the individual's prior consent and the data is handled for purposes other than outputting the response, the business may be in breach of the APPI. It should therefore confirm that the AI provider will not use the personal data for machine learning, among other things.
3. When the PPC must be told and individuals notified
Where a leak, loss, damage or other security incident involving personal data occurs that falls within the categories set by PPC rules as likely to cause serious harm to individuals' rights and interests, the business must report it to the PPC (APPI, Art. 26(1)) and notify the individuals (Art. 26(2)). The categories are (Enforcement Rules of the APPI, Art. 7):
- A leak, or possible leak, of personal data including sensitive personal information (such as medical history)
- A leak, or possible leak, of personal data whose misuse could cause financial loss (such as credit card numbers)
- A leak, or possible leak, caused by an act that may have been carried out for a wrongful purpose
- A leak, or possible leak, involving more than 1,000 individuals
Whether entering data into an AI service is a "leak" at all depends on how the service works and the contract terms. It is important to organise the facts and decide early whether a report is required.
4. Effect on trade secrets and partners' confidential information
A trade secret protected by the Unfair Competition Prevention Act is technical or business information useful for business activities that is kept secret and is not publicly known (Unfair Competition Prevention Act, Art. 2(6)). A single input does not necessarily mean the protection is lost. But if staff are tacitly allowed to enter information and there are no rules, it may become harder to argue that the information was "kept secret".
If information received from a business partner was entered, check the confidentiality agreement with that partner. Clauses prohibiting use for other purposes or disclosure to third parties, and requiring notice of a leak, will set the deadline and method for your response.
5. Dealing with the employee
Start by hearing the employee's account and recording what happened. Disciplinary action requires a basis in the work rules; where an employer regularly employing 10 or more workers has rules on sanctions, their types and severity must be set out in the work rules (Labor Standards Act, Art. 89(ix)). Even then, a disciplinary measure is void as an abuse of rights if, in light of the nature and manner of the conduct and other circumstances, it lacks objectively reasonable grounds and is not considered appropriate in general social terms (Labor Contracts Act, Art. 15).
If the employee was using the tool to get work done faster and the company had no rules on AI use, a severe sanction is more likely to be found disproportionate. Where the company was slow to set rules, it is usually more realistic to communicate rules company-wide and prevent recurrence before considering discipline.
6. Using overseas AI services
Many generative AI services are run by overseas providers. Providing personal data to a third party in a foreign country generally requires the individual's prior consent to that cross-border provision (APPI, Art. 28(1)). When seeking consent, the business must give the individual information such as the data protection regime of that country (Art. 28(2)).
Whether entering data into an AI service counts as "provision", or only as outsourcing or the company's own use, depends on how the provider handles the input. Japanese subsidiaries of foreign companies that use an AI tool chosen by the parent should also check the contract terms from the perspective of the APPI.
7. What to put in your internal AI rules
The core of prevention is making clear what is allowed. Including at least the following makes it easier for staff to decide what to do:
- Permitted services and contract type: whether only corporate plans contracted by the company may be used, or personal accounts are allowed.
- Information that must not be entered: personal data on customers or staff, sensitive personal information, information received from business partners, unpublished financial information, source code and so on.
- Exceptions: whose approval is needed when input is genuinely necessary.
- Handling of output: assume it may contain errors and have a person check it before it leaves the company. The PPC's alert also points to the risk that responses contain inaccurate content.
- Consequences of breach: how the rules relate to the disciplinary provisions in the work rules.
Rules work only if they are communicated. Training all staff and keeping records of acknowledgement also makes it easier to show that information was "kept secret" for trade secret purposes.
What you can do now
- Record what was entered, when, which service and type of account, and the training settings.
- Where possible, delete the input history in the service and switch off use for training.
- If personal data was included, check whether the incident falls within any of the four categories in Article 7 of the Rules.
- If a partner's information was involved, check the notice obligations and deadlines in the confidentiality agreement.
- Set rules on which AI services may be used, what information must not be entered and how approval is given.
- Communicate the rules to all staff and keep records of training and acknowledgements.
When to talk to a lawyer
- Sensitive personal information, or customer data on more than 1,000 people, was entered.
- You cannot tell whether a report to the PPC or notice to individuals is required.
- A business partner has asked for an explanation of how its information was handled, or for damages.
- You are considering disciplining the employee, but your work rules are vague.
- You plan to use an AI tool introduced by your overseas parent company in Japan as it is.
Frequently asked questions
Q. Is entering information into ChatGPT a data leak in itself?
It depends on the service and settings. The assessment differs between a setting where inputs may be used for the provider's training and a corporate plan whose contract excludes training. Check the terms and settings first and keep a record.
Q. Must we always report to the PPC?
A report is mandatory only where a leak of personal data falls within the categories in Article 7 of the Rules (APPI, Art. 26(1)). Typical cases are those involving sensitive personal information or more than 1,000 individuals. Even where no report is required, internal records and measures to prevent recurrence are still needed.
Q. Can we dismiss the employee for disciplinary reasons?
Discipline requires a basis in the work rules, and a sanction that is not proportionate to the nature of the conduct and the circumstances is void (Labor Contracts Act, Art. 15). Where the company had no rules on AI use, or the employee used it for work, a severe sanction is unlikely to be upheld.
Q. Should we ban AI use altogether?
A total ban can drive staff to use AI secretly on personal devices. Clearly identifying permitted services and information that must not be entered, and managing use through corporate contracts, is often a better way to protect information.
Q. Is it fine to enter personal data into a corporate plan?
Even with a corporate plan, you need to check that the use is within the purpose of use (APPI, Art. 18) and that security measures are adequate (Art. 23). Where the provider is overseas, also check how the rules on provision to third parties in foreign countries (Art. 28) apply.
FREE CONSULTATION
Corporate Law: your first 30 minutes are free
- ・Handled by Yusuke Abe, admitted in Japan and New York
- ・Online (Zoom etc.) or at our office in Otemachi, Tokyo
- ・In English or Japanese
This article provides general legal information and does not constitute legal advice. For specific legal issues, please consult with a qualified attorney.
Related Questions
Q. Can a director in Japan be held personally liable for damages? Liability to the company and to third parties, and the mechanisms that limit it
Q. Is a shareholders' agreement legally binding in Japan? How far it reaches, and which terms belong in the articles of incorporation
Q. How enforceable are post-employment non-compete clauses in Japan? How courts assess validity, what employers can do on breach, and points to note for foreign parent company templates
Q. Can a company terminate a business contract before it expires in Japan? Termination by agreement, under a termination clause and for breach, and what to check before you terminate
Q. How do you register a trademark in Japan? Official fees, the first-to-file rule and what overseas companies need to know
Related Articles
Can You Commercially Use Generative AI Output in Japan? Copyright Article 30-4 and Output-Stage Risk
A guide for businesses on whether generative AI output can be used commercially in Japan: Article 30-4 of the Copyright Act and its proviso, the similarity and reliance test at the output stage, how training data affects reliance, when AI-generated material is protected by copyright, injunctions, damages and criminal penalties, and steps you can take now.
Entering Japan as a Foreign Company: Representative Office, Branch or Subsidiary, and the Governing Law of Your Contracts
How a foreign company can enter Japan through a representative office, a branch or a subsidiary, explained against the Companies Act: the ban on continuous transactions before registration, the representative in Japan, KK versus GK, governing law and jurisdiction clauses, Foreign Exchange Act filings, and the October 2025 changes to the Business Manager status of residence.
Free Tools
Free 30-minute initial consultation
Yusuke Abe, admitted in Japan and New York, will handle your matter — in Japanese or English, online or in person.