SAKURA Law Office | Legal Update by Managing Partner Kenshiro Michishita
September 24, 2026
SAKURA Law Office Managing Partner Kenshiro Michishita
Executive Summary
Entering trade secrets or confidential information into generative AI is not categorically prohibited. However, doing so without first determining whether the AI service may use the information for training, improvement, retention or disclosure to third parties; whether the provider is bound by confidentiality obligations; and whether an NDA or other contract restricts third-party disclosure or use outside the agreed purpose can create material risk to statutory trade-secret protection and contractual confidentiality obligations. METI’s March 2025 Management Guidelines for Trade Secrets make clear that the secrecy management requirement is not automatically lost merely because confidential information is generated or output within an internal generative-AI environment, while also recognizing that disclosure to a third-party AI service provider may, depending on the circumstances, undermine that requirement. Corporate practice should therefore focus not on the binary question of whether information was “entered into AI,” but on who processes what information, for what purpose, under what contractual framework, and with what technical settings and access controls.
SAKURA Law Office is pleased to publish the fourth installment of the Legal Update series by Managing Partner Kenshiro Michishita, entitled “May Companies Enter Trade Secrets and Confidential Information into Generative AI? — Key Practical Issues in NDAs, Trade Secret Management and AI Service Agreements (2026).”
As generative AI becomes an ordinary corporate tool, employees are increasingly entering unpublished business plans, customer lists, pricing strategies, source code, research and development information, product specifications, board materials, M&A documents, litigation materials, information received from third parties under non-disclosure agreements and other confidential information into AI systems for purposes such as summarization, contract review, code generation, data analysis and meeting-note preparation.
It would be inaccurate to say categorically that “trade secrets must never be entered into generative AI.” It would be equally inaccurate to say that “enterprise AI is safe, so confidential information may be entered freely.” The legal analysis depends on the terms governing the relevant AI service, whether data is retained or used for training, the confidentiality obligations imposed on the provider, disclosures to third parties and subprocessors, the user company’s information-management measures, the nature of the information, the terms of any NDA and the surrounding facts.
METI’s Handbook for Protection of Confidential Information, revised in February 2024, cautions against unintended disclosure through careless entry of information into external generative-AI services and emphasizes the need for rules limiting use to approved AI systems and approved categories of information. METI’s Management Guidelines for Trade Secrets, revised in March 2025, went further by adding specific analysis of generative AI in relation to the “secrecy management” requirement for trade-secret protection.
In February 2025, METI also published its Contract Checklist for AI Utilization and Development, which highlights the need to determine in advance whether inputs may be used to train foundation models, whether those inputs fall within the contractually protected data set, and whether they may be provided to third parties. In April 2026, METI published Technology Leakage Prevention Guidance, Version 2, emphasizing that internal rules should be continuously reviewed in light of changes in business practices, including the use of generative AI.
This Legal Update examines, as of September 24, 2026, the legal analysis applicable when companies enter trade secrets or confidential information into generative AI. Drawing on Japan’s Unfair Competition Prevention Act, METI guidance on trade secrets and public materials concerning AI contracting practice, it addresses NDAs, AI service agreements, internal rules and incident response in the order in which these issues typically arise in practice.
1. “Trade Secrets” and “Confidential Information” Are Not the Same Thing
The first step in evaluating whether information may be entered into generative AI is to distinguish between a “trade secret” and “confidential information.” The terms are often used interchangeably in business, but they are not legally identical.
To qualify as a “trade secret” under Japan’s Unfair Competition Prevention Act, information must be managed as secret, must constitute useful technical or business information for commercial activities, and must not be publicly known. These are commonly referred to as the secrecy management, utility and non-publicity requirements.
Not every item a company labels “Confidential” or otherwise treats as sensitive necessarily satisfies the statutory requirements for a trade secret. Conversely, information may be protected by an NDA or other contractual confidentiality obligation even if it does not meet the statutory definition of a trade secret.
Accordingly, the decision whether information may be entered into AI should not turn solely on whether the information qualifies as a trade secret. Companies must identify the distinct legal basis protecting each category of information, including statutory trade secrets, contractually confidential information, personal information, sensitive technology from an economic-security perspective, and access credentials.
2. The Secrecy Management Requirement Turns on Whether the Company’s Intention to Maintain Secrecy Is Recognizable, Not on Formalities Alone
Among the three statutory requirements for trade secrets, the secrecy management requirement is particularly important in the context of generative AI.
METI’s Management Guidelines for Trade Secrets explain that the purpose of the secrecy management requirement is to identify, within a company, the information that is intended to be kept secret and thereby ensure that employees and other relevant persons can recognize the company’s intention to maintain confidentiality.
The law does not invariably require the highest imaginable level of security for every piece of information. What matters is that, taking into account the company’s size, operations and the nature of the information, measures such as access restrictions, confidentiality markings, internal rules, training, contracts and system controls create a state in which it is reasonably recognizable that the information is to be treated as secret.
Now that generative AI has become a routine business tool, secrecy management must address not only traditional risks such as removing files or sending them by email, but also whether particular information may be entered into external AI systems. If a company has no AI-use rules and permits employees, over an extended period, to enter important information into arbitrary external AI services, the reasonableness of the company’s overall trade-secret management framework may itself become an issue.
3. Confidential Information Does Not Automatically Lose Trade-Secret Protection Merely Because It Is Generated or Output Within an Internal AI Environment
The March 2025 revision of METI’s Management Guidelines for Trade Secrets provides an important clarification concerning generative AI and trade secrets.
The Guidelines explain that where information managed as secret within a company is used as training data for generative AI and is later generated or output as AI-generated material within the company’s own secrecy-management unit, the mere fact of generation or output does not, by itself, defeat the secrecy management requirement, provided that the information continues to be managed as secret within that unit.
In other words, using generative AI in business does not automatically cause information to lose legal protection as a trade secret. Where a closed environment, an internal model or another appropriately access-controlled enterprise AI environment is used, and both inputs and outputs remain subject to proper internal confidentiality management, AI use and trade-secret protection can coexist.
The decisive issue is not the label attached to the technology. It is how far the information moves, who can access it, and what controls continue to apply.
4. Entering Trade Secrets into an External Generative-AI Provider Raises a Different Question
The analysis changes where the same information is provided to a third-party generative-AI service provider outside the company.
METI’s Management Guidelines for Trade Secrets expressly recognize that where information is provided to an external third party, such as a generative-AI service provider, the secrecy management requirement may in some circumstances be denied. METI materials addressing generative AI likewise caution that entering trade secrets into an external AI service operated by a provider that is not bound by confidentiality obligations may result in the information falling outside the scope of trade-secret protection.
Before trade-secret information is entered into an external AI service, companies should therefore confirm, at a minimum, the purposes for which the provider may use the input; whether it may be used for training or model improvement; whether the input is retained; whether human access is possible; whether subprocessors may access it; whether it may be disclosed to third parties; and whether the provider is contractually bound by confidentiality obligations.
The analysis should not stop at the phrase “enterprise plan.” Data-use terms may differ materially among consumer services, enterprise plans, API-based implementations and particular administrative configurations, even under the same AI brand.
5. “Not Used for Training” Is Not the Same as “Never Leaves the Company”
One of the first questions companies ask when evaluating an AI service is whether submitted data is used to train the provider’s model. That is important, but it is not the end of the analysis.
Even where inputs are not used for model training, they may be retained in logs for a period of time, accessible to personnel for abuse monitoring or incident response, processed by subprocessors, or transferred to servers outside Japan.
For trade secrets and NDAs, companies must review the full data lifecycle rather than focusing only on a training opt-out. This includes submission, transmission, retention, processing, access, subprocessors, backups, deletion and post-termination handling. The question is whether contractual and technical safeguards proportionate to the sensitivity of the information exist throughout that lifecycle.
Conversely, if an external service operates in a dedicated environment in which inputs are not used for training, provider access is tightly limited by contract and technology, and confidentiality and deletion obligations are established, the fact that the system is externally provided does not necessarily justify a blanket prohibition on its use.
6. Information Received Under an NDA Requires a Separate Contractual Analysis in Addition to Trade-Secret Law
A recurring source of risk in practice is information received under an NDA from a business partner, portfolio company, seller, buyer, joint-research partner or other third party.
NDAs often prohibit use of confidential information outside the agreed purpose, restrict disclosure to third parties without consent, limit the categories of persons who may access the information, and regulate subcontracting, reproduction, cross-border transfers, return and destruction.
Companies must therefore examine the language of the NDA to determine whether entering the information into an external AI service falls within the permitted contractual “use,” whether transmission to the AI provider may constitute a “disclosure” or “provision,” whether the provider can be treated as an authorized contractor or service provider of the recipient, and whether prior consent is required.
The answer may also depend on whether the AI provider is bound by confidentiality obligations, may use the data for its own purposes, or permits human access. The technical fact that data is transmitted to an AI system does not automatically establish a contractual breach in every case, but information should not be entered without first understanding the obligations imposed by the NDA.
7. Many Existing NDAs Were Not Drafted with Generative AI in Mind
Traditional NDAs frequently anticipate disclosure to directors, officers, employees, lawyers, accountants and financial institutions, but do not expressly address processing by cloud-service providers or generative-AI vendors.
Companies that expect to use generative AI in material matters should therefore revisit their NDA templates and define the extent to which information-processing services, cloud providers, AI services and subprocessors may be used.
From the recipient’s perspective, a blanket ban on every cloud or AI service that is reasonably necessary to perform the work may be operationally burdensome. From the discloser’s perspective, however, it may be unacceptable for sensitive information to be submitted, at the recipient’s sole discretion, to unspecified external AI services.
A more practical approach is to design NDA provisions according to the sensitivity of the information, addressing approved services, confidentiality and security obligations imposed on service providers, restrictions on training and secondary use, and prior consent or notice where appropriate.
8. AI Service Agreements Should Confirm That Inputs Fall Within the Contractually Protected Data Set
In an AI service agreement, the contractual definition of data submitted to the service is critical.
METI’s Contract Checklist for AI Utilization and Development provides a framework for separately reviewing the identification, provision, use, external disclosure, rights allocation and processing results associated with data supplied by an AI user to a service provider.
In practice, companies should confirm that the relevant contractual definitions of “Customer Data,” “Input” or similar protected categories cover not only prompt text, but also uploaded files, images, audio, internal data connected through retrieval-augmented generation (RAG), data transmitted via API, logs and metadata.
If certain categories of data fall outside the contractual definition and the provider retains broad usage rights or no confidentiality obligations apply to them, the provider may be permitted to use information in ways the customer did not anticipate.
9. Secondary Use and Disclosure of Inputs Containing Trade Secrets Must Be Reviewed Carefully
Where trade-secret information may be entered into AI, two of the most important contractual issues are secondary use by the service provider and disclosure outside the provider.
Companies should determine whether the provider processes inputs solely to deliver the contracted service or may also use them for model training, quality improvement, analytics, new-service development or other independent purposes. If secondary use is permitted, the scope of that use, anonymization or aggregation, and any opt-out mechanism should also be reviewed.
Where input data may be provided to affiliates, subprocessors, infrastructure providers or other third parties, companies should review the scope of recipients, applicable confidentiality obligations, security standards, cross-border processing and responsibility in the event of an incident.
An AI-contract review should therefore extend beyond a high-level statement that “your data is not used for training.” Terms of service, data-processing terms, privacy policies, security documentation and administrative settings should be reviewed together.
10. Source Code, API Keys, Passwords and Other Credentials Require Particular Caution
Generative AI is widely used in software development, but source code may embody company-specific algorithms, security architecture, internal APIs, data structures and other trade secrets.
If source code contains API keys, access tokens, passwords, private keys or other credentials, entering it into an external AI service can create not only a trade-secret issue but an immediate risk of unauthorized access and information compromise.
As a general rule, credentials should not be entered into AI systems. If an accidental submission occurs, the company should not rely solely on a deletion request; the affected key or token should be revoked and reissued, passwords changed, and other credential-rotation measures implemented promptly.
Where development teams use AI coding assistants, it is prudent to adopt controls separate from those applicable to general chat-based AI, including repository-level access restrictions, training-use settings, retention of corporate code, and the scope of external transmission.
11. M&A, Financing, Litigation and Board Materials Should Be Treated as High-Risk AI Use Cases
M&A, financing, litigation, internal investigations and board matters often involve unpublished information capable of materially affecting corporate value.
Such materials may contain not only trade secrets, but also material nonpublic information of listed companies, personal information, sensitive communications with legal counsel, contractually confidential information and information relevant to economic security.
Where material from important matters is to be processed by external AI, companies should apply a higher approval threshold than for ordinary business use. Depending on the circumstances, this may mean limiting use to a dedicated enterprise environment, requiring prior legal and information-security approval, or entering only anonymized or abstracted portions rather than the original documents.
The convenience of AI should not lead companies to upload an entire transaction data room, complete board materials or full litigation files into external AI systems without defined limits and controls.
12. AI Agents and RAG Require Management of Access Rights, Not Merely Rules About What Users May “Input”
With AI agents and RAG systems, employees may not manually paste confidential information into a prompt at all. The AI may instead connect directly to cloud storage, internal databases, email, CRM systems or code repositories.
In that environment, rules about what employees may type into a prompt are no longer sufficient. The central issue becomes access governance: which folders, databases, mailboxes, customer records and other data sets the AI is permitted to access.
The principle of least privilege should apply. AI systems should receive only the access necessary to perform the relevant function, while highly sensitive data sets may require specific approval, human review or exclusion from the AI connection altogether.
AI agents also create risks such as prompt injection, in which malicious instructions embedded in external content may cause an AI system to disclose confidential information. Trade-secret management and cybersecurity therefore need to be designed as an integrated framework.
13. Companies Must Also Exercise Caution When AI Outputs Appear to Contain Another Company’s Trade Secrets
If generative-AI output contains another party’s trade secrets or Shared Data with Limited Access, the output is not necessarily free for the recipient to use.
METI materials concerning AI recognize that where another party’s trade secret or Shared Data with Limited Access is contained in a trained model or AI-generated output, subsequent use or disclosure may, depending on the circumstances, constitute use or disclosure of the underlying protected information.
If AI output appears to contain a competitor’s internal documents, unpublished designs, customer lists or comparable information, the company should not immediately use it in the business. The source of the information, its confidential nature and the circumstances of acquisition should be investigated.
The assumption that “AI produced it, so it is free to use” is dangerous not only in copyright law, but also in the law of trade secrets and protected data.
14. Generative AI May Also Implicate Japan’s “Shared Data with Limited Access” Regime
Some data exchanged between businesses is not necessarily non-public in the same sense as a trade secret, but is accumulated and managed electronically in substantial volume and provided, under defined conditions, to specified recipients.
Japan’s Unfair Competition Prevention Act provides a separate framework protecting qualifying “Shared Data with Limited Access” against certain acts of improper acquisition, use and disclosure.
In AI development, joint research, data-collaboration projects and platform businesses, companies may input or train systems on data that could fall within this regime even where the data does not qualify as a trade secret.
Accordingly, where data has been received from another company, the analysis should not end with the conclusion that “it is not a trade secret because it is publicly known.” Companies should also consider whether the data may constitute Shared Data with Limited Access or be subject to contractual restrictions on use.
15. Sending Sensitive Technology to an Overseas AI Service May Also Raise Technology-Transfer Issues under FEFTA
Trade secrets and confidential information may include technical information subject to Japan’s security export-control regime under the Foreign Exchange and Foreign Trade Act (FEFTA).
METI explains that certain transactions involving the provision of controlled technology to a foreign country or to a nonresident, as well as taking technology out of Japan for the purpose of providing it abroad, may require a license under FEFTA’s service-transaction rules. Transmission by email and comparable electronic means can constitute the provision of technology.
For cloud services, METI distinguishes between mere storage and substantive technology provision. Where the sole purpose is storage and the service provider does not view, obtain or use the technical information, use of a server located abroad does not, by itself, necessarily constitute a regulated service transaction. By contrast, where the user knows that the provider or another party may view, obtain or use the controlled technology, the transaction may be regarded in substance as one intended to provide technology.
Accordingly, where design information, manufacturing technology, software or other sensitive technical information potentially subject to export controls is entered into an overseas AI service, companies should coordinate not only with trade-secret and contracting personnel but also with the function responsible for security export control.
16. METI’s 2026 Technology Leakage Prevention Guidance, Version 2 Calls for Continuous Updating of Internal Rules in Light of Generative AI
On April 27, 2026, METI published Technology Leakage Prevention Guidance, Version 2.
The Guidance expands measures addressing technology leakage associated with overseas production bases and personnel to include risks arising in joint research and procurement coordination. It also emphasizes that technology-management rules and training should be continuously reviewed as business practices change, including by adding or updating rules for the use of generative AI.
Entering information into generative AI has therefore moved beyond the category of a narrow IT-use issue. It should be treated as part of corporate technology-leakage prevention, economic security, intellectual-property strategy and supply-chain management.
In joint research, outsourced manufacturing, technology licensing and overseas operations in particular, a company cannot protect its information through its own internal rules alone if the counterparty is using AI. AI-use conditions must be designed across the supply chain through contract terms and counterparty management.
17. An “Information Classification × Approved AI” Matrix Is an Effective Tool for Corporate AI Policies
A generative-AI policy will not function in practice if it says only that “confidential information must not be entered.” Employees need rules that allow them to determine what is permitted in an actual business situation.
One practical approach is to classify information into categories such as public information, ordinary internal information, confidential internal information, trade secrets or highly sensitive information, third-party confidential information and credentials, and to classify AI environments into public AI services, company-approved enterprise AI and dedicated closed environments. Each combination can then be designated as permitted, permitted only after anonymization, subject to approval, or prohibited.
For high-risk use cases such as M&A, HR, research and development, source code and export-controlled technology, separate use-case-specific approval rules may be appropriate in addition to the general information classification.
The purpose of the policy should not be to discourage legitimate AI use. It should define, in concrete terms, the circumstances in which employees can use AI safely.
18. Immediate Response When Trade Secrets or Confidential Information Are Entered into AI by Mistake
If an employee accidentally enters important information into an external AI service and conceals the incident, the company loses valuable time in which mitigation may be possible. The incident-response framework should therefore encourage immediate reporting.
The initial assessment should identify the information entered and its sensitivity, the AI service, account and plan used, the time of submission, chat or API logs, training-use settings, retention periods, and the potential access of the service provider and any subprocessors.
Where possible, the relevant chat or file should be deleted, the provider asked to delete or refrain from using the information, relevant account settings changed, and access suspended if appropriate. If API keys, passwords or similar credentials were included, they should be revoked and reissued immediately.
Where the information was received from a third party under an NDA or similar agreement, the company should review any contractual notification, consultation or incident-reporting obligations. If personal information is involved, separate analysis may also be required under Japan’s data-breach rules.
The company should also determine, technically and contractually, whether the information was actually viewed by or output to a third party and whether training or other secondary use may have occurred. Legal, information-security and business teams should then determine the appropriate response.
19. Confidential-Information Governance for AI Must Be Designed Jointly by Legal, IT/Security and the Business
Trade-secret and NDA issues cannot be solved by reading the law alone. A contract may prohibit training use, but the control framework will fail if employees use a different AI service through personal accounts. Conversely, sophisticated technical controls will not establish effective secrecy management if employees do not understand what information the company expects them to treat as secret.
Legal should define statutory trade-secret requirements and contractual obligations. IT and security should govern approved services, configurations, logs and access rights. Business teams should identify actual use cases and the information genuinely required to perform the work.
Only when those functions operate together can a company capture the benefits of AI while protecting confidential information.
As corporate AI use expands, confidential-information governance must evolve from a system designed merely to “lock information away” into one that permits information to be used safely under controlled conditions.
20. Frequently Asked Questions
Q1. Does entering a company trade secret into ChatGPT or another generative-AI service automatically cause it to cease being a trade secret?
No. Trade-secret protection is not automatically lost in every case. METI has clarified that where information remains properly managed as secret within an internal AI environment, the fact that the information is generated or output by AI does not, by itself, negate the secrecy management requirement. By contrast, where information is provided to an external AI provider that is not bound by confidentiality obligations, secrecy management may in some circumstances be denied. The service terms, configuration, access rights and actual information-management practices must be examined.
Q2. If an enterprise AI service states that customer data is not used for training, is it safe to enter trade secrets?
The absence of training use is important, but it is not sufficient. Companies should also review retention, provider and subprocessor access, cross-border processing, incident notification, deletion and contractual confidentiality obligations. For information received from a third party, the NDA must also be reviewed to determine whether use of an external service is permitted at all.
Q3. May we ask an AI service to summarize a contract received from a business partner under an NDA?
It depends on both the NDA and the AI service. The company should determine whether the use is within the contractual purpose, whether transmission to the provider could constitute third-party disclosure or subcontracting, whether the provider is bound by confidentiality obligations, and whether the input may be used for secondary purposes. If those points cannot be confirmed, alternatives include anonymizing or abstracting the information or using an approved closed environment.
Q4. May we submit source code to AI for review?
If the source code constitutes the company’s trade secret or confidential information received from a third party, the terms and configuration of the AI environment must be reviewed. API keys, passwords, private keys and other credentials should generally not be entered into external AI. If credentials are submitted accidentally, they should be revoked and reissued immediately.
Q5. May we use output that appears to contain a competitor’s internal information?
Caution is required. If the output may contain another party’s trade secret or Shared Data with Limited Access, its use or disclosure may create issues under the Unfair Competition Prevention Act. Important business or product-development decisions should not rely on such information until its source and legal status have been investigated.
Q6. Can entering technical information into an overseas AI service create an issue under FEFTA?
Yes, potentially. If the information constitutes controlled technology and the transaction amounts to a provision of technology to a foreign country or nonresident, a license may be required. The analysis differs from mere foreign-server storage: where the service provider may view, obtain or use the information, the transaction may be treated as a technology provision and should be reviewed from an export-control perspective.
Q7. Is it enough for a generative-AI policy to say only that confidential information must not be entered?
Usually not. “Confidential information” is too broad and vague for employees to apply consistently. A more effective policy combines information classifications with categories of approved AI and specifies, for each combination, whether use is permitted, requires anonymization, requires approval, or is prohibited.
Q8. What should we do first if information subject to an NDA is accidentally entered into generative AI?
The company should immediately identify the information, the service, account and settings used, and whether the data may be retained or used for training. Deletion and a request that the provider not use the information should be considered promptly. At the same time, the company should review notification or other obligations under the NDA and investigate the possibility of third-party access or secondary use. If credentials are involved, they should be revoked and reissued immediately.
21. The SAKURA Law Office Perspective on Generative AI, Trade Secrets and NDAs
It is easy to say that companies should never enter confidential information into AI. But that statement alone does not answer the practical challenge of protecting corporate value while using AI in real business operations.
The more important task is to understand the nature and sensitivity of the company’s information and design the conditions under which each category may be processed in each AI environment. Statutory trade-secret management, NDAs with third parties, AI service agreements, access controls, cybersecurity and security export controls should not be treated as isolated regimes. They must operate together as an integrated information-governance framework.
Overly restrictive controls aimed solely at protecting information may drive legitimate AI use outside the company’s governance structure and increase the risk of shadow AI. At the other extreme, prioritizing convenience while neglecting contracts and information management may allow a single submission to jeopardize years of accumulated technology, know-how or business trust.
Corporate legal practice should avoid both extremes. The objective is to define, with precision, the range within which AI can be used safely in light of the company’s business, information and risk profile.
22. About the Kenshiro Michishita Legal Update Series
SAKURA Law Office continues to publish the Legal Update series on the profile page of Managing Partner Kenshiro Michishita, addressing legal issues of practical importance to companies and to the business community.
The first installment addressed corporate use of generative AI and AI governance in Japan. The second examined whether companies may enter customer and personal information into ChatGPT and other generative-AI services under Japan’s data-protection framework. The third addressed generative AI and copyright, including AI training, AI-generated content and commercial use.
This fourth installment addresses trade secrets, confidential information and NDAs, areas in which enterprise AI use can create particularly significant business harm. It examines secrecy management, submission of information to external AI providers, AI service agreements, third-party confidential information, AI agents, technology leakage and incident response in a single practical framework.
Future Legal Updates will continue to address AI service agreements, internal AI policies, AI agents, M&A, cross-border transactions, Web3 and digital assets, crisis management, corporate misconduct and other issues of importance to corporate legal practice, with close attention to both law and real-world business operations.
23. How SAKURA Law Office Can Assist with Generative AI, Trade Secrets and NDAs
SAKURA Law Office advises companies on trade-secret, confidential-information, NDA and AI-service-contract issues arising from the introduction and use of generative AI.
Our work may include assessing whether a company may enter its trade secrets or confidential information into a proposed AI environment; reviewing whether information received under an NDA may be processed with AI; reviewing provisions on training use, confidentiality, third-party disclosure, subprocessors and deletion in AI service agreements; updating NDA and confidentiality clauses for generative-AI use; incorporating trade secrets and third-party confidential information into internal AI policies; designing information controls for AI agents and RAG systems connected to internal data; assessing legal risks when sensitive technology is processed through overseas AI services; and responding to accidental submissions or other information-security incidents.
Companies may consult us before the relevant AI service or operating model has been finalized. By reviewing the information involved, the proposed AI environment, applicable contract terms and the contemplated workflow, we can help structure the contractual, policy, technical and organizational controls, as well as the incident-response framework, appropriate to the use case.
24. Generative AI and Trade-Secret Legal Inquiries
For advice concerning the submission of trade secrets or confidential information to generative AI, NDAs, AI service agreements, trade-secret management, internal AI policies, AI agents, RAG, technology leakage or other AI and IT-law matters, please contact SAKURA Law Office.
For corporate clients, our support includes not only disputes and crisis response after an information incident, but also pre-deployment legal reviews, preparation and negotiation of NDAs and AI contracts, internal policies, information-classification frameworks, approval workflows and continuing AI and information governance.
When contacting us, noting that your inquiry concerns “Generative AI and Trade Secrets” will help us direct the matter to the appropriate lawyer promptly.
SAKURA Law Office
Managing Partner: Kenshiro Michishita
4F, Ark Hills South Tower, 1-4-5 Roppongi, Minato-ku, Tokyo 106-0032, Japan
Tel: +81-3-6910-0692
https://sakura-lawyers.jp/en/
Principal Japanese Sources
e-Gov, Unfair Competition Prevention Act of Japan
METI, Contract Checklist for AI Utilization and Development
Ministry of Internal Affairs and Communications / METI, AI Guidelines for Business (Version 1.2)
METI, Technology Leakage Prevention Guidance, Version 2
METI, Security Export Control Guidance [Introductory Edition]
METI, Technology-Related Q&A (Cloud Computing Services and Related Matters)
Written and supervised by SAKURA Law Office, Managing Partner Kenshiro Michishita
This article provides general legal information based on Japan’s Unfair Competition Prevention Act, METI’s Management Guidelines for Trade Secrets, Handbook for Protection of Confidential Information, Contract Checklist for AI Utilization and Development, Technology Leakage Prevention Guidance and other publicly available materials as of September 24, 2026. The effect of submitting information to generative AI on the secrecy management requirement for trade-secret protection or on obligations under an NDA or other contract depends on the terms and technical specifications of the AI service, the applicable settings, the nature of the information, the possibility of third-party access, the wording of the relevant agreements and other circumstances. AI-service terms, data-use conditions and technical specifications may change over time. Specific matters should be reviewed in light of the latest laws, judicial decisions, administrative materials, contractual relationships and technical facts.