1. Business operator and person responsible for the protection of personal information

We are a godo kaisha (a limited liability company under Japanese law), and the person responsible for the handling of personal information is our representative member. The name of the business operator under Article 32, paragraph 1, item 1 of the Act on the Protection of Personal Information (APPI, Japan) (the "APPI") is as set out in the table above. Our address will be provided without delay upon request to support@phrasy.net, as permitted by that item.

We are not a covered business operator of any accredited personal information protection organization.

2. Scope

This Policy applies to the handling of personal information in the iOS / Android app "Phrasy", the servers (API) of the Service, and the web pages that we publish. The web pages are https://phrasy.net/privacy / https://phrasy.net/terms / https://phrasy.net/support, and the product introduction pages (https://phrasy.net/, and the language-specific pages /en , /ko and /zh-tw; together, the "LP").

The statements about the Android version (Google Play, Play Integrity, FCM and so on) apply from the time we start providing the Android version. Until then, only the statements about the iOS version apply.

The conditions for using the Service are set out in the Terms of Service (https://phrasy.net/terms). This Policy forms part of the Terms of Service. The disclosure of transaction conditions for the Paid Plan is posted on the page of disclosures based on the Act on Specified Commercial Transactions (Japan) (https://phrasy.net/legal/commerce).

We have not included the European Economic Area, the United Kingdom or Switzerland among the App Store countries and regions where the app is distributed, and we do not offer or advertise the Service to residents of those regions. The statements in this Policy concerning the General Data Protection Regulation (GDPR) are included in case we offer the Service in those regions in the future, or in case the GDPR applies by operation of law.

This Policy does not apply to the handling of personal information on third-party sites or services linked from the Service.

3. Information we collect

The information we collect is as follows. We do not collect any other information (the device's location data (such as GPS), contacts, photos, calendar, health information, payment card information, date of birth, or telephone number). Even where the sign-in provider passes us the URL of a profile image, we do not collect the image itself.

We collect "Subscription information" only in respect of users who purchase the Paid Plan. Item (b) of "Device information for preventing the free lesson from being reused" applies from the version of the app that incorporates DeviceCheck.

By default the Service can be used with an anonymous account, in which case we do not collect names, email addresses or telephone numbers. Unless you sign in with Apple, Google or a code sent by email, the only user identifier that we hold is a random ID. Registration by email address and password is not offered, and we do not collect passwords (sign-in by email is done by entering the code that is sent to you).

About sensitive personal information. Where what a user enters or says on their own (Chat, conversations, reports, disputes, free-text answers to surveys and so on) contains sensitive personal information such as medical history, creed, race or criminal record (Article 2, paragraph 3 of the APPI; this includes sensitive information [민감정보] in Korea), we acquire and handle it to the extent necessary to provide the Service. Users outside Korea are deemed to consent to this by entering or saying it. Users in Korea (those whose device region is Korea or whose display language is Korean) are asked, separately and by a separate checkbox on the consent screen, for separate consent to the handling of sensitive information (Article 23, paragraph 1, item 1 of the Personal Information Protection Act of Korea). The purpose is to provide the Service, the item is the sensitive information contained in what the user enters or says on their own, and the retention period is as stated in Section 9. This checkbox is optional, and you can still use the Service as usual if you do not consent. Even then, please do not enter sensitive information. We do not ask you to enter such information. Please do not enter it. The special categories of personal data under Article 6 of the Personal Data Protection Act of Taiwan (the six types: medical records, medical treatment, genetics, sex life, health examinations and criminal records) are not covered by the consent above. We do not ask for this information and do not use it for providing the Service. Please do not talk about it in conversations.

On the Android version (applicable from the time we start providing it), to prevent reuse of the free lesson we use Google's Play Integrity API (including its Device Recall feature) in place of DeviceCheck. We do not receive an identifier that identifies the device; we receive only the result of the device assessment that Google returns and the mark that the free lesson has been used on that device. When we send notifications from the server on Android, they are delivered through Google's Firebase Cloud Messaging (FCM). We collect the notification destination as the FCM token and the Firebase installation ID that FCM uses to tell devices apart, both only for a device on which you have allowed notifications. We store the record of what we sent in the same way as on iOS.

If you use Chat, we store the messages you send to a character and the character's replies. We also store what the AI remembers about you (short sentences made from what you said; up to 12 lines per user per character, up to 160 characters per line). In addition, we store the plans settled in Chat (the date, the time, who takes part and what happens; up to 12 per user, not counted in the 12 lines above), and delete each plan 14 days after its date has passed. It can include what you talked about, such as your tastes, everyday topics or plans. We instruct the AI (for plans too) not to remember health, money, home address or exact location, contact details, religion, politics, sex or romance, self-harm or suicidal thoughts, or anything you ask it to forget, but this depends on the AI's judgment and we do not guarantee that such content is never included in what it remembers or in the plans. Chat is available only to users who have signed in.

When you send a dispute about the AI's judgment in a review, we store the type and text (optional) of the dispute, a copy of the original evaluation, and the result of the AI evaluator's re-check.

When you answer the survey about why you purchased or cancelled, we store the reason you chose, an optional free-text comment, and your subscription status and plan at the time. You can continue without answering either. Please do not write personal information such as your name or contact details in the free-text comment.

We collect the notification destination (push token) only from a device on which you have allowed notifications, in order to deliver notifications from the server (Section 16). We do not collect it if you do not allow notifications or if you turn notifications off in the app, and we revoke any destination already collected.

We do not collect the advertising identifier (IDFA), and we do not link your activity with other companies' apps or websites. However, only for iOS users who have allowed it both on the explanation screen in the app and in the device's tracking permission, the Meta SDK sends the advertising identifier (IDFA) and other information directly to Meta Platforms, Inc. (Sections 7 and 15).

4. Purposes of use

We use the information we collect only to the extent necessary to achieve the following purposes. Where we use it beyond these purposes, we obtain the individual's consent in advance.

• Providing the Service — generating the conversation partner's replies, reading out audio, displaying subtitles and native-language translations, and giving feedback on conversations and guidance for rephrasing
• Evaluating and recording learning — judging whether a phrase was used successfully, building review schedules, and displaying records and progress
• Operating the mechanisms that support continuation — scores, streaks, League ranking and weekly settlement
• Applying and managing the Paid Plan — checking the purchase status (by querying RevenueCat), determining the trial and paid periods, switching the features available, storing transaction records, moving the ownership of a subscription and recording that move when you restore a purchase on a different account, and responding to inquiries about refunds and cancellation. In order to understand profitability, we keep a copy of purchases in our database. Whether the Paid Plan is available is decided by querying RevenueCat, and only when we cannot reach RevenueCat do we decide using this copy instead
• Understanding practice usage — applying the limit beyond the free range, and understanding the number of practices and the situation in which the limit is reached (the practice allowance records and the records of reaching the limit; Section 3)
• Preventing reuse of the free lesson — to keep the condition that the free lesson is available once per device, we use device information (Section 3: the hash of the installation identifier, and Apple's DeviceCheck). We do not use this information for advertising or combine it with other companies' data
• Signing in and verifying the person — signing in with Apple, Google or a code sent by email (including sending the email that contains the code), showing the email address on the "Profile > Account" screen, determining and deleting guest accounts that are no longer used, and revoking the Apple sign-in token when an account is deleted
• Contacting the individual — confirming who sent a report or a dispute, replying to reports, and giving notice of important changes
• Improving quality — identifying and fixing defects (including looking at the photo of the screen attached to a report in order to check what happened and fix it), improving materials and features, and understanding response speed and failure rates
• Analyzing usage, and tailoring features and guidance — understanding usage for each answer in the first-run guide (Section 3), and tailoring features and guidance according to those answers
• Reusing the conversation partner's replies — to use, among the replies generated by the conversation partner (AI), short ones that contain no numbers, proper nouns, contact details or personal circumstances, unchanged as to their text and the audio in which they were read out, as the partner's reply in other users' conversations in the same scene (Section 13). As examples for the matching that selects which reply to use, we store the transcription of your utterance immediately preceding that reply. The text and audio that are reused do not include your utterances, recordings or identifiers
• Tallying visits to the LP — to understand, in a form that does not identify individuals, the number of visitors to the LP, where they come from and the rate at which they press the App Store button, and use this to improve advertisements and the LP (Section 15)
• Measuring the effect of advertising — to count installs that came through an Apple ad by campaign and so on, and to measure which ads led to use (Sections 3 and 15). In addition, only for iOS users who have allowed it, we provide app events and similar information to Meta for measuring and delivering advertisements (Section 7)
• Preventing misuse and ensuring safety — applying the usage limits, and detecting and responding to unauthorized access and abuse (including the check by Cloudflare Turnstile that prevents abuse of guest creation and of sending codes by email, and the management of sessions using the connection records that the login mechanism stores), and identifying use for review and testing of the app so as to exclude it from other users' screens such as the League (the "Record of the determination of the type of use" in Section 3)
• Safety check of nicknames and the sentences that are candidates for reuse, and of Chat replies and Chat messages — to check with OpenAI's Moderation API, before they are shown, sent or taken in, that the nicknames shown to other users, the reply sentences that may be used in other users' conversations, and the sentences of the AI-written Chat replies and Chat messages contain no inappropriate expressions. If the check cannot be completed, a Chat partner's reply or Chat message is not shown or sent (Section 7)
• Safety check of conversations — to check with OpenAI's Moderation API whether the AI-written replies of the conversation partner in lesson conversations and Chat, and the sentences you type when Chat returns a prepared reply (when it could not read your message, or when the daily limit is reached), contain self-harm or sexual content. If an AI reply does, it is not shown and is replaced with a prepared one. If, when Chat returns a prepared reply, your sentence contains self-harm, the prepared reply is a caring one. This check is not perfect: it can miss things and can detect things by mistake. It is not intended as emergency response or monitoring (Section 7)
• Statistical analysis — understanding usage, retention and costs in a form that does not identify individuals (including identifying use for review and testing of the app and excluding it from the statistics), and improving lessons and features by using, only for the purpose of statistics, the learning records and the reasons chosen in surveys that we keep after an account is deleted with the user ID replaced by a random ID (Section 9)
• Analysis and operation by the operator — for analyzing usage of the Service, investigating defects and quality, responding to reports and disputes, and understanding costs, the operator may have the information in Section 3 (such as conversation content, transcripts, learning records, usage logs, and report and dispute content) read by external AI services (Claude by Anthropic, and ChatGPT, Codex and the API by OpenAI; Section 7), which summarize, classify and analyze it. We use them only with settings under which what is read is not used to train that service's models
• Providing Chat — to create replies in text conversations with characters, and to weave what the AI remembers into later Chat conversations and into the first reply of lessons with the same character (what it remembers is stored so that conversations can continue naturally; deleting your learning data or your account deletes all of it; Section 12)
• Reviewing disputes about the AI's judgment — to answer your dispute and for a person at the operator to check errors in the AI's judgment and fix quality. The content of a dispute is not reflected in your learning records, points, or review schedule
• Tallying surveys — to tally reasons for purchase and cancellation and use them to improve pricing and guidance
• Performing obligations under applicable law, and responding to the exercise of rights and to disputes
• Sending notifications ── to deliver, through the Apple Push Notification service (APNs) or, on Android, Google's Firebase Cloud Messaging (FCM), notices when your streak is about to be broken, messages from a Chat partner (the one message the partner wrote) and the start, update and end of a Live Activity (a feature that shows the time left until the deadline on the Lock Screen and elsewhere). We keep a record of what we sent so as not to send the same notification twice

We do not use the information we collect for evaluating a user's creditworthiness, for decisions on employment, recruitment, insurance or lending, or for behaviorally targeted advertising. However, the provision to Meta in Section 7 (only for those who have allowed it) is used to help improve the delivery of our advertisements, and is also used by Meta, in accordance with its own terms and policies, for the personalization of advertisements and similar purposes. We do not make important decisions about an individual automatically (without human judgment). Feedback and evaluations of conversations are generated by AI, but these are information provided as a reference for learning, and are not decisions that affect the user's legal status or rights.

Table for users in Taiwan (Article 8 of the Personal Data Protection Act). The correspondence between our purposes of use and the numbers of the specific purposes and categories of personal data in the Ministry of Justice's "Specific Purposes and Categories of Personal Data under the Personal Data Protection Act" (個人資料保護法之特定目的及個人資料之類別) is as follows.

5. Bases for processing

Where the EU / UK General Data Protection Regulation (GDPR) or another law that requires the legal basis for processing to be stated applies, we process personal data on the following bases.

The handling of audio in the Service is for the purpose of reading what has been spoken (English-language phrases) and evaluating it for the sake of learning. We do not handle it for the purpose of uniquely identifying a specific individual from the characteristics of their voice. Accordingly, we do not treat audio as a special category of personal data (biometric data) within the meaning of GDPR Article 9. We will not repurpose this processing in the future for speaker identification or voice authentication.

We add the following pursuant to Article 8 of the Personal Data Protection Act (Taiwan). Even where you do not make audio recordings (do not allow the microphone), you can use the Service with keyboard input, but the features relating to the evaluation of pronunciation and speech are not available. Where you do not register a nickname, you are shown as "Guest". Where the information necessary to store your profile and learning records cannot be obtained, features such as displaying records, review schedules and participation in the League are not available. Not answering the questions in the first-run guide places no restriction on your use of the Service. Where you do not sign in with Apple, Google or a code sent by email, you can use the Service up to the one free lesson, but you cannot use the free trial or the Paid Plan. However, a lesson that we release free of charge to all users for a stated period (such as a special episode) can be practiced during that period without signing in, and does not count toward the one free lesson.

6. Use of generative AI, and non-use as training data

The conversation partner, the feedback on and evaluation of conversations, and the translations in the Service are created automatically using generative AI. The conversation partner is AI and is not a human. The partner's voice, and the voice that reads the lessons aloud, are each a voice synthesized by AI, and are not human voices. For these purposes, we use the APIs provided by OpenAI, L.L.C. (United States), Theai, Inc. (Inworld AI; United States) and TypeSafe AI, Inc. (United States), as follows.

• Transcribing audio (OpenAI; we send the recording)
• Generating the conversation partner's replies (OpenAI; we send the flow of the conversation and the learner's utterances)
• Synthesizing the speech of the partner's lines, of the model sentences in the feedback (the sentences in which the evaluation rewrites what you said into natural English), and of the lessons' fixed sentences (Key Phrases, examples and the lines of the scenes) (Inworld; we send only the English text of the reply created by AI, the English model sentence, or the fixed sentences in the lessons (which do not contain anything related to an individual), and the specification of the voice and speed. We do not send your recordings. A model sentence is a rewrite of what you said, so it may contain words you said, such as a name or place. The AI's reply may contain them as well)
• Translating into the native language, giving feedback on and evaluation of conversations, and guidance for rephrasing (OpenAI; we send the content of the conversation)
• Translating the body of reports (OpenAI; so that the operator can read them; we send the body)
• Chat replies and updates of what the AI remembers (OpenAI; we send the messages you sent in Chat with the date and time of each message in your local time, and what has already been remembered, including settled plans)
• Re-checking a dispute (OpenAI; we send the text of the dispute, the original evaluation, and the transcript of that conversation)
• Selecting candidates for reusing replies, and judging the confirmation before a Key Phrase is rephrased (TypeSafe AI; we send a summary of the scene, the flow of the conversation, the learner's utterance, and the reply that is a candidate for matching. We do not send the transcription of another user's utterance)
• Checking nicknames, the sentences that are candidates for reuse, and the AI-written replies and messages in Chat for inappropriate content (OpenAI's Moderation API; we send only the nickname, the utterance and reply sentences that are candidates for reuse, and the sentence of a Chat partner's reply or of a Chat message that the character sends on its own; we do not send email addresses or other personal information. If the check cannot be completed, the nickname or candidate is not adopted, and a Chat partner's reply or Chat message is not shown or sent, for safety)
• Safety check of conversations (OpenAI's Moderation API; in lesson conversations and Chat we send, one sentence at a time, the sentences of the AI-written conversation-partner replies and the sentences you type when Chat returns a prepared reply (when it could not read your message, or when the daily limit is reached). We do not attach your user ID or email address. The sentence may contain words you said, such as a name. As a result of this check, we do not record the category that was hit or the sentence. If the check cannot be completed, the lesson conversation continues, and in Chat the reply is not shown, as before)
• Judging whether a reply typed in silent practice gets across (TypeSafe AI; we send the description of the scene, the partner's line, what the learner wanted to say, the target phrase, and the English the user typed)

The speech of the partner's lines, of the model sentences in the feedback (the sentences in which the evaluation rewrites what you said into natural English), and of the lessons' fixed sentences is synthesized by Inworld only. What we send is only the English text of the reply created by AI, the English model sentence, and the fixed sentences in the lessons (which do not contain anything related to an individual). However, because both the reply and the model sentence are created in response to what the user said, words the user said (such as a name or place) may be contained in them and sent to Inworld.

Inworld's terms of use (https://inworld.ai/terms) provide that it will not train its generally available models on materials that are not publicly available (except for feedback and for materials that are flagged for review on suspicion of a violation of the terms). Inworld's data processing addendum (https://inworld.ai/data-processing-addendum) is incorporated into its service agreement, and Inworld acts as a processor on our instructions. We do not send Inworld your recordings or personal information such as an email address. A model sentence is a rewrite of what you said, so it may contain words you said (such as a name or place).

We ourselves do not use what users send (recordings, transcripts, conversation content, the text of reports, disputes and Chat, and so on) to train or fine-tune any AI model. We have no mechanism for training our own model, and we do not ask our subcontractors to train on it. How each AI provider that we subcontract to handles it differs by provider, as follows (to the extent we have been able to confirm; each company's policy may change).

OpenAI. Under OpenAI's policy for its API (https://developers.openai.com/api/docs/guides/your-data), content that we send through OpenAI's API is not used to train or improve OpenAI's models unless the business opts in. We have not opted in to data sharing. OpenAI may retain the content sent to and received from the API for up to 30 days for the purpose of abuse monitoring (whether it is retained differs by feature; except where a longer retention period is required by law, or where retention is necessary to address conduct that harms the service). We attach a setting that prevents the response from being stored by OpenAI (store: false) to the calls that generate conversation, translate and evaluate. The call for transcribing recordings does not have this setting.

Selecting candidates for reusing replies, judging the confirmation before a Key Phrase is rephrased, and judging whether a reply typed in silent practice gets across are performed through the API of TypeSafe AI, Inc. (United States). TypeSafe AI's privacy policy (https://typesafe.ai/legal/privacy-policy; updated November 19, 2025) states that it will not train or fine-tune its models on prompts or other input. It does not publish a data processing agreement, standard contractual clauses or a list of sub-processors, and we have not entered into a separate data processing agreement with it. As to retention, its policy says only that it keeps personal data for as long as reasonably necessary to provide the services, and does not state a specific period. Therefore, that what we send to TypeSafe AI is not used for training rests on the statement in its published policy, not on an agreement we have made. We send this data after obtaining your consent, on the explanation and consent screen, to provision to a third party in a foreign country. What we send is only the text needed for the judgment (the flow of the conversation and the transcription of the user's utterance; in silent practice, the description of the scene, the partner's line, what the learner wanted to say, the target phrase and the English the user typed); we do not send recordings, names or email addresses. Before sending, we mask email addresses, phone numbers, URLs and runs of five or more digits, and we do not send the user's ID, name or email address.

The replies, translations, feedback and evaluations generated by AI, and the voice synthesized by AI, may contain errors or unnatural points. They are provided as a reference for learning, and their accuracy is not warranted (please see the Terms of Service).

7. Provision to third parties, and subcontracting

We do not sell or provide personal information to third parties without the individual's consent. To the extent necessary to provide the Service, we subcontract processing to the following companies. Each of them acts as a party that processes data on our instructions (a subcontractor / processor), and, under our contracts, each company's terms and settings, we require that they not use users' information for their own purposes. We have entered into agreements on the handling of personal data (data processing agreements) with these companies, or have agreed to the terms of such agreements that each of them publishes. TypeSafe AI, Inc. is an exception to this description. It does not publish a data processing agreement, and we have not entered into a separate data processing agreement with it. It states in its own privacy policy that it does not train or fine-tune its models on input; on that basis, and after obtaining your consent on the explanation and consent screen, we send it only the information needed for the judgment. Cloudflare, Inc. is also an exception to this description, because for improving Turnstile itself it handles information as a controller on its own account (see "About Cloudflare (Turnstile)" below). In addition, under Cloudflare's self-serve terms of service, its data processing agreement is incorporated only where the data includes personal data of EU or UK data subjects or information covered by the CCPA. That agreement therefore does not apply to our users' information, and Cloudflare's processing as a processor is governed by its terms of service (the provisions on data processing and security) and the Turnstile privacy addendum. Separately, the operator's use of external AI services for analysis (including Anthropic, PBC in the table) is not treated as part of this subcontracting (see "About the use of external AI services in analysis by the operator" below). Under Japan's Act on the Protection of Personal Information, this is carried out as provision to a third party and provision to a third party in a foreign country, after obtaining your consent on the explanation and consent screen.

Please see each company's own policy for how it handles data.

• OpenAI: https://openai.com/policies/privacy-policy/ / API data handling: https://developers.openai.com/api/docs/guides/your-data / List of sub-processors: https://openai.com/policies/sub-processor-list/
• Supabase: https://supabase.com/privacy / Data processing agreement: https://supabase.com/legal/dpa
• Vercel: https://vercel.com/legal/privacy-policy / Data processing agreement: https://vercel.com/legal/dpa
• GitHub: https://docs.github.com/site-policy/privacy-policies/github-general-privacy-statement / List of sub-processors: https://docs.github.com/en/site-policy/privacy-policies/github-subprocessors
• Expo: https://expo.dev/privacy / List of sub-processors: https://expo.dev/privacy/subprocessors
• Slack: https://slack.com/trust/privacy/privacy-policy / List of sub-processors: https://slack.com/slack-subprocessors
• RevenueCat: https://www.revenuecat.com/privacy/ / Data processing agreement: https://www.revenuecat.com/dpa/
• Inworld: https://inworld.ai/privacy / Data processing agreement: https://inworld.ai/data-processing-addendum
• TypeSafe AI: https://typesafe.ai/legal/privacy-policy / Terms: https://typesafe.ai/legal/terms
• Zoho (ZeptoMail): https://www.zoho.com/privacy.html / List of sub-processors: https://www.zoho.com/privacy/sub-processors.html
• Cloudflare: https://www.cloudflare.com/privacypolicy/ / Terms of service: https://www.cloudflare.com/terms/ / Turnstile privacy addendum: https://www.cloudflare.com/turnstile-privacy-policy/ / Data processing agreement: https://www.cloudflare.com/cloudflare-customer-dpa/ / List of sub-processors: https://www.cloudflare.com/gdpr/subprocessors/
• Google (FCM): https://policies.google.com/privacy / Firebase privacy and security: https://firebase.google.com/support/privacy / Data processing terms: https://firebase.google.com/terms/data-processing-terms
• Anthropic (where used for analysis by the operator): https://www.anthropic.com/legal/privacy / List of sub-processors: https://www.anthropic.com/subprocessors

About Cloudflare (Turnstile). Before a guest account is created and before a code is sent by email, the app runs Cloudflare's Turnstile (a mechanism that tells humans from bots) in a web view that you do not see. Cloudflare's Turnstile Privacy Addendum (https://www.cloudflare.com/turnstile-privacy-policy/; updated June 18, 2025) provides that Cloudflare processes signals such as the device's IP address, the TLS characteristics and the User-Agent as a processor acting on the instructions of its customer (us) for detecting and blocking bots, and as a controller on its own account for improving Turnstile itself. Our processing as its customer is limited to the purposes in Section 3, and our instructions do not extend to Cloudflare's own purposes. The login mechanism (Supabase) sends the token with the result of the check to Cloudflare in order to verify it. The addendum does not state Cloudflare's retention period, and we do not know it.

About sub-processing. Each company may use sub-processors that process data under its own contract. We permit this on the basis of each company's data processing agreement, and for the companies that publish a list of their sub-processors we have given the link above. TypeSafe AI does not publish a list of its sub-processors. For the companies for which no link to a list is given (Supabase, Vercel, RevenueCat and Inworld), the sub-processors follow the terms of that company's data processing agreement and policies. Inquiries about sub-processors may be sent to the contact point in Section 18 or to the contact point of each company in Section 8.

About the use of external AI services in analysis by the operator. For analyzing usage of the Service, investigating defects and quality, responding to reports and disputes, and understanding costs, the operator may have the information in Section 3 (such as conversation content, transcripts, learning records, usage logs, and report and dispute content) read by external AI services (Claude by Anthropic, PBC (United States), and ChatGPT, Codex and the API by OpenAI, L.L.C. (United States)), which summarize, classify and analyze it. We use them only after setting them so that what is read is not used to train that service's models. We limit the information that is read to the scope necessary for the analysis. Claude by Anthropic may be used under a consumer plan. In that case, we are not under Anthropic's data processing agreement for business customers, and we use it only after setting it so that what is read is not used for training. Under Anthropic's terms, retention under this setting is 30 days.

About notifying the operator of disputes (the "Challenge" feature in the app). When a dispute is made, we send the text of the dispute, the AI's re-check result, the original evaluation, the full transcript of that conversation, the first 8 characters of the user's internal identifier, and the nickname and email address of the person who sent the dispute to the operator's business messaging tool (Section 7), and a person at the operator checks it. The check is made to fix errors in the AI's judgment, and its outcome does not rewrite your learning records, points, or review schedule. The transcript includes what you said. When you delete your learning data or your account, we delete the posts of reports and disputes from the business messaging tool (Section 9).

About the operator-only connection (MCP). Through this connection, the operator (only a person authenticated as an administrator) may view records in our database (including the information in Section 3, which can include personal data) in read-only mode from connectors of external AI services such as Anthropic's Claude or OpenAI's assistant. The purpose is the same as the analysis by the operator above; we use these services only under settings in which what is loaded is not used to train the provider's models, and every view is recorded in the log of operator actions. General users cannot use this connection.

Provision for advertising measurement and delivery. Only if you have allowed it both on the explanation screen in the app and in the device's tracking permission, the Meta SDK sends the following to Meta Platforms, Inc. (United States): the app events (install, launch, sign-up, completion of the first lesson), the SDK identifier, the advertising identifier (IDFA), advertising campaign identifiers, device information (model, OS, screen, language, region, time zone, mobile carrier, number of CPU cores and so on), the app's identifier and version, the URL schemes the app registers, the IP address, the state of the tracking permission, the advertising identifier collection setting and the data processing setting. Meta uses this for measuring and delivering advertisements and, in accordance with its own terms and policies, also associates it with the users of its services and uses it for the personalization of features and content (including advertisements and recommendations) on and off its products, for safety, for research and for improving its products. This is not a delegation by us. Under its terms, Meta retains event data for at most 2 years. This provision applies from the version of the app that includes the Meta SDK, from the time we turn it on. The features and prices of the Service do not change if you do not allow it. You can withdraw your permission at any time in "Settings". When you withdraw it, we do not send events or the advertising identifier. The SDK is not run from the next launch. How information sent before you withdrew is handled depends on Meta's policy. You can change Meta's ad personalization in the advertisement settings of each Meta service (Facebook's guide: https://www.facebook.com/help/568137493302217 ). A US industry group's guide, http://www.aboutads.info/choices, is also available, but it may not be usable in your country.

About the company with which the user contracts directly. App distribution, Sign in with Apple, purchases and payments (App Store billing), DeviceCheck, and the measurement of advertising effectiveness (SKAdNetwork and the attribution lookup for Apple's advertisements) are carried out by Apple Inc. on the basis of its own terms and privacy policy (https://www.apple.com/legal/privacy/). Apple is not a subcontractor that processes data on our instructions, but an independent business operator that bears its own responsibility for these matters. What we receive from Apple is the email address when the user uses Sign in with Apple (a relay address where the user chooses "Hide My Email"), the result of a DeviceCheck query (whether the mark showing that the free lesson has been used on that device is present), and the result of the attribution lookup for Apple's advertisements (Section 15). What we send to Apple is the token for querying and updating DeviceCheck and the token for the attribution lookup, each obtained by the app from the device. We do not receive from DeviceCheck an identifier that identifies the device. Verification of purchases and management of the subscription status are entrusted to RevenueCat (the table above). We never receive payment card information. Also, when a user who signed in with Sign in with Apple deletes the account, we ask Apple to revoke the sign-in token, in accordance with Apple's mechanism. We do not store the Apple token: at the time of confirming the deletion, we show Apple's confirmation screen again, send the authorization code obtained to Apple on the spot, exchange it for the token, and revoke it immediately.

About push notifications. Notifications from the server are delivered to your device through the Apple Push Notification service (APNs) of Apple Inc. Apple is not a subcontractor that processes on our instructions; it is an independent company that is responsible for APNs under its own terms and privacy policy (https://www.apple.com/legal/privacy/). What we pass to Apple is only the notification destination (the push token and the Live Activity token) and the content of the notification (for a streak notice, the number of days of your streak and fixed wording; for a Chat message, the one message the Chat partner wrote; and the values used to display the time left in a Live Activity). We do not pass recordings or your email address.

Sign-in with Google is carried out by Google LLC on the basis of its own terms and privacy policy (https://policies.google.com/privacy). Sign-in takes place through our authentication mechanism (Supabase), and we receive from Google, to the extent the user has agreed to in signing in, basic information such as the email address. Google is not our subcontractor, but an independent business operator that bears its own responsibility for sign-in.

On the Android version (applicable from the time we start providing it), Google LLC carries out app distribution, purchases and payments (Google Play billing), the Play Integrity API, on the basis of its own terms and privacy policy (https://policies.google.com/privacy). Google is not our subcontractor but an independent business operator that is itself responsible for these. What we pass to Google is the token for Play Integrity API queries; we do not pass recordings or your email address. We entrust the delivery of notifications to Android not to an independent business operator but to Google, as set out in the row of the table above. Verification of purchases and management of the subscription status are entrusted to RevenueCat, as on iOS.

Apart from the above, we may provide personal information without consent only where this is based on laws and regulations, where there is an urgent need to protect the life, body or property of a person, where it is necessary to cooperate with a national agency or the like in performing duties prescribed by law, and where this accompanies a business succession.

Where you participate in the League, your nickname, Nico's outfit, and your score and rank for that week are shown to the other users in the same group (up to 30 people) (Section 13). This is a display that is part of the functionality of the Service, and is distinguished from provision to third parties.

8. Provision to third parties in foreign countries (cross-border transfers)

All of the subcontractors in Section 7 are located outside Japan (in the United States and India). On the explanation and consent screen shown before use of the Service begins, we obtain consent to the fact that this includes provision to third parties in foreign countries (Article 28, paragraph 1 of the APPI). Pursuant to paragraph 2 of that Article, we set out reference information below.

For users in Korea, we make the overseas transfer under Article 28-8, paragraph 1, item 3 of the Personal Information Protection Act (Korea), disclosing in this Section the matters in paragraph 2 of that Article (the items transferred, the countries, the timing and method, the recipients and their contact points, the purposes and retention periods, and how to refuse and the effect of refusing). The contact point of each recipient of the overseas transfer (the department responsible for the handling of personal information) is as follows.

For users in Korea, the provision to Meta Platforms, Inc. (United States) for advertising measurement and delivery is not a subcontracting of processing, and is made only where the separate consent of the data subject (Article 28-8, paragraph 1, item 1 of the Korean Personal Information Protection Act) has been obtained.

• Items transferred: app events (install, launch, sign-up, completion of the first lesson), SDK identifier, advertising identifier (IDFA), advertising campaign identifiers, device information (model, OS, screen, language, region, time zone, mobile carrier, number of CPU cores and so on), the app's identifier and version, the URL schemes the app registers, IP address, state of the tracking permission, advertising identifier collection setting, data processing setting
• Country, time and method: United States / when the app is used after the consent and the device's tracking permission / sent over the network through the SDK
• Recipient and contact: Meta Platforms, Inc. (1 Meta Way, Menlo Park, CA 94025, USA). Contact: Meta Platforms, Inc. ATTN: Privacy Operations, 1 Meta Way, Menlo Park, CA 94025
• Purpose and retention and use period: measuring and delivering advertisements, association with the users of its services, personalization of features and content (including advertisements and recommendations) on and off its products, safety, research and improving its products / under its terms, event data is retained for at most 2 years
• Method, procedure and effect of refusal: you may refuse by not consenting, or withdraw at any time in "Settings"; the features and prices of the Service do not change if you refuse.

Apple Inc. and Google LLC (both U.S. companies) are, as stated in Section 7, not our subcontractors but independent business operators (except for Google's delivery of notifications on Android); however, information is exchanged with them for DeviceCheck queries and the attribution lookup for Apple's advertisements (in each case, a token obtained by the app from the device is sent to Apple) and for sign-in (we receive basic information from Google). Each handles information under its own policy (Apple: https://www.apple.com/legal/privacy/ / Google: https://policies.google.com/privacy).

To APNs (Apple Inc., United States) we send the notification destination and the content of the notification (the number of days of your streak and fixed wording, or the one message the Chat partner wrote). Apple is not our subcontractor but an independent company, and its handling follows Apple's policy (https://www.apple.com/legal/privacy/).

We add the following pursuant to Article 28-8 of the Personal Information Protection Act (Korea). Users may refuse the overseas transfers described above. However, the transcription of recordings, the generation of replies and the storage of data are all core to the Service, and we cannot provide the Service without these transfers. Accordingly, if you refuse the transfers, please either not begin using the Service, or withdraw your consent from "Profile > Data & privacy" in the app and delete your account. If you refuse, the learning records accumulated up to that point become unavailable.

Where the GDPR applies, for transfers by us outside the European Economic Area we ensure safeguards through each company's data processing agreement, which includes the Standard Contractual Clauses adopted by the European Commission (GDPR Article 46(2)(c)). Japan is recognized by the European Commission as a country that ensures an adequate level of protection.

9. Retention periods

The database has daily backups for recovery from failures, and we keep the most recent 7 days (the specification of Supabase's Pro plan; https://supabase.com/docs/guides/platform/backups). A copy within them remains for up to 7 days after the deletion operation. This copy is not used for any purpose other than recovery from failures, and disappears with the passage of that period. Where we carry out a recovery, we run the deletion process again after the recovery.

When you request deletion of your account, the Service becomes unavailable at that point. The physical deletion of personal data takes place within 30 days of the request. The deletion process runs once a day and targets those whose request is 29 days old or more (so that it does not exceed the 30-day limit). If you wish to cancel, please contact the contact point in Section 18 before the deletion process begins (until 29 days have passed since the request). Once it begins, the deletion cannot be cancelled.

In the deletion process, for usage logs and records of external API use we keep the rows themselves but delete the user ID, the installation identifier and the per-launch identifier. This is in order to preserve the statistics of daily usage counts and costs. These rows never contained names, email addresses, the text of utterances, or the characters of the phrases studied. Once the identifiers have been deleted, we do not link these rows to a specific individual, and we retain them without any time limit only for the purpose of statistics. The same applies when you use “Delete all learning data”: at that point we keep the rows of usage logs and records of external API use and delete the user ID, the installation identifier and the per-launch identifier (we do not wait for an account deletion).

In the deletion process, we delete the information that identifies you: the user ID, nickname, sign-in email address and authentication records, the user's record on the RevenueCat side, device identifiers, push tokens and the record of what we sent, answers in the first-run guide, conversation transcripts, recordings, the text of reports, disputes and Chat, the photos of the screen attached to reports, the free text of surveys, League results, the list of blocks, and the record of the determination of the type of use. Learning records (lesson completions, evaluation judgements, the mastery state and review records for each phrase, scores and streaks, and practice allowance records) and the reasons chosen in surveys are kept after the user ID is replaced with a new random ID that cannot be turned back into the original, without keeping any table that links it to the original ID. Items that may contain free text are deleted from the kept records as well. The kept records contain no attributes of you other than the display language. We use them, without a fixed period, only for the purpose of statistics, to improve lessons and features (see the next paragraph for how we treat them). If you do not want your learning records kept either, use “Delete all learning data” before deleting your account (Section 12).

How we treat the learning records kept after an account is deleted. The kept records are in a form where the user ID has been replaced with a new random ID that cannot be turned back into the original, and the free text and device identifiers have been removed (we do not keep a table that links it to the original ID). Because they remain as a history of rows for each person, we do not claim that they are information that cannot identify an individual (anonymously processed information); we treat them as information that may be regarded as personal information (or pseudonymously processed information), as follows. (1) The items in the kept records are lesson completions, evaluation judgements, the state of each phrase, review records, scores, streaks, practice allowance records, the reasons chosen in surveys, and the display language. (2) We use them only for statistics to improve lessons and features. (3) We do not try to identify the person behind a record, and we do not match it against other information such as subscription records, device information or email addresses. (4) We do not contact anyone using these records. (5) We do not provide them to third parties (except for storing them in a subcontractor's storage area). (6) We apply the security measures in Section 10 to these records as well. Since we keep no table linking to the original ID, even when we receive a request we cannot confirm which record belongs to the person making it. We cannot respond to requests for disclosure or deletion with respect to records that cannot be confirmed as the person's own.

Procedure and method of destruction. Personal information whose retention period has passed, and personal information whose deletion has been requested, is destroyed by deleting rows in the database and deleting files in the storage area, in a way that cannot be restored. We do not store it on paper media. Copies in backups disappear with the passage of time, as stated above.

Records of operator actions are kept even after deletion, because it is necessary to retain when, by whom, on which user, and what was done. However, the notes and free-text reasons written by the operator are removed at the time of deletion.

10. Measures taken for security management

Pursuant to Article 23 of the APPI, we take the following measures (published pursuant to Article 32, paragraph 1, item 4 of that Act and Article 10, item 1 of the Order for Enforcement of that Act).

Where a leak, loss, damage or other incident involving personal information occurs and there is a significant risk of harm to the rights and interests of individuals, we report it to the Personal Information Protection Commission pursuant to Article 26 of the APPI and notify the individuals. Where the GDPR applies, we notify the supervisory authority in principle within 72 hours, in accordance with Articles 33 and 34 of that Regulation.

11. Users' rights and how to exercise them

Users have the following rights with respect to their own information. Many of them can be exercised immediately from within the app (Section 12). For the others, please make a request to the contact point.

Please send requests to support@phrasy.net, together with information that allows us to confirm that the request comes from the individual themselves. After we have confirmed identity, we will respond in principle within two weeks, and at the latest within the period prescribed by law. There is no fee. Where we cannot comply with a request, we will tell you the reason. Where we comply with a request for deletion, the same two-stage procedure as when deletion is requested from the app (Section 12) applies, and the scope of what is deleted is the same. Deletion of an account is requested from within the app; if you wish to cancel during the 30-day grace period, please contact the contact point.

If you cannot open the app or have disposed of your device, you can request deletion of your account by the procedure "Request account deletion from the web" on the support page (https://phrasy.net/support).

Where you are not satisfied with our response, you may lodge a complaint with a supervisory authority. Japan: Personal Information Protection Commission (https://www.ppc.go.jp/). Korea: Personal Information Protection Commission (https://www.pipc.go.kr/), as well as the Personal Information Infringement Report Center (Korea Internet & Security Agency [KISA]; https://privacy.kisa.or.kr/ / 118 without a country code), the Personal Information Dispute Mediation Committee (https://www.kopico.go.kr/ / 1833-6972), the Supreme Prosecutors' Office (https://www.spo.go.kr/ / 1301 without a country code) and the National Police Agency (https://cyberbureau.police.go.kr/ / 182 without a country code). Taiwan: Preparatory Office of the Personal Data Protection Commission (個人資料保護委員會籌備處; https://www.pdpc.gov.tw/). European Economic Area / United Kingdom: the supervisory authority of your place of residence or work (GDPR Article 77).

12. What you can do from within the app

You can carry out the following operations at any time from "Profile > Data & privacy" in the app. There is no procedure and no review; they are carried out the moment you tap.

Deletion of an account takes place in two stages: the stopping of use at the time of the request, and physical deletion within 30 days. You can cancel until the deletion process begins (until 29 days have passed since the request). If you wish to cancel, please contact the contact point in Section 18. Once the process begins, it cannot be undone. Any unsent records remaining on the device at the time of deletion are discarded without being sent. Even where the deletion of some recording files fails, those files are automatically deleted after 72 hours from storage, and at the latest within 4 days (Section 9). The physical deletion is carried out as a single unit in the database, so if it fails partway nothing changes (you are never left in a half-deleted state).

If you cannot open the app, you can request deletion of your account by email, following the procedure on the support page (https://phrasy.net/support). After we have confirmed that you are the person concerned, we will in principle reply within two weeks to say that we have received the request, and the deletion is carried out in the same two stages as above.

Where a user who has been using the Service as a Guest signs in to a different, existing account that already has records and switches to it, the Guest's records are not carried over. As with account deletion, the information that identifies the user (conversation transcripts, feedback, profile and so on) is deleted, and the learning records are kept with the user ID replaced by a random ID (Section 9; only the usage logs and the records of permission for advertising measurement are reassigned to the account signed in to; the subscription records are kept as set out in Section 9). The app notifies you when you switch. Where you sign in as a new account that has no records yet, the Guest's records are carried over to the new account.

Our Supabase project is dedicated to Phrasy and is not shared with any other app. When the deletion of an account completes, we also delete the account registered with our authentication mechanism (Supabase), including the email address.

13. Information visible to other users (the League, and the conversation partner's replies)

The Service has a "League", in which up to 30 people learning during the same period compete on scores each week. While you are participating, the following information is shown to the other users in the same group.

• Nickname (shown as "Guest" where it is not set)
• Nico's outfit (the look of our companion character "Nico"; you choose from 4 types. We do not handle photographs)
• The score earned that week, and your rank within the group

The content of conversations, recordings, feedback, email addresses and device information are never visible to other users. You may choose your nickname freely. Please do not put in your nickname a name or contact detail that you do not want others to know. An account that we judge to be an unauthorized use, or use for review and testing of the app, may not be shown on other users' screens such as the League standings (Section 3).

If you turn off "Join the league" in "Profile > League", you are not assigned to a new group thereafter. For a week in which you are already participating, the display remains until that week ends, because removing it would change the ranks of the other people in the same group after the fact.

In the League, you can block other users. When you block someone, their nickname and Nico's outfit are hidden in your table (their rank and score remain). The person you block is not told, and their own screen does not change. For this we store the ID of the user you blocked and the time, and keep it until you unblock them or delete your account. When an account is deleted, both the blocks you made and the blocks made against you are deleted.

Replies generated by the conversation partner (AI) in conversations with you may be used unchanged, as the partner's reply (the text and the audio in which it was read out), in the conversations of other users who made a similar utterance in the same scene. Conversely, a partner's reply that appears in your conversation may also have been generated in another user's conversation. Reuse is limited to short replies that contain no numbers, proper nouns, contact details or personal circumstances, and a program takes in only those that meet these conditions. The operator checks the replies in use by sampling, and stops using any that reflect personal circumstances. Your utterances themselves, recordings, nickname and any other information that can identify you are never visible to other users through this mechanism.

14. Use by minors

The Service is intended for persons aged 13 or older. Persons under 13 may not use the Service. We do not intentionally collect personal information from persons under 13. Where it comes to light that we have collected information from a person under 13, we delete it promptly.

If you are under 18 (under 19 if you use the Service from Korea), please use the Service only after obtaining the consent of a parent or other legal representative. Where a person is found not to have the capacity to judge the consequences of consenting to the handling of personal information, consent must be obtained from their legal representative.

The provision to Meta in Section 7 is intended for persons aged 18 or older. Persons under 18 should not allow it.

If you are using the Service from Korea, under Article 22-2 of the Personal Information Protection Act (Korea) the consent of a legal representative is required in order to handle the personal information of persons under 14. We ask for confirmation that you are 14 or older on the consent screen, and therefore in Korea persons under 14 may not use the Service.

If you are using the Service from the European Economic Area, under GDPR Article 8 and the laws of your country of residence, the consent of a person under 16 (or under such other age as the Member State has prescribed, being 13 or above) requires the consent or authorization of a legal representative.

Where a legal representative wishes to request the disclosure or deletion of a child's information, please contact support@phrasy.net.

15. Cookies, tracking and advertising

The app for the Service itself does not use cookies. It does not incorporate third-party analytics SDKs. In the iOS app, we incorporate the Meta SDK for advertising measurement and delivery. This applies from the version of the app that includes it, from the time we turn it on, and the SDK runs only for those who have allowed it both on the explanation screen in the app and in the device's tracking permission, and sends information to Meta Platforms, Inc. (United States) (Section 7). Apart from that, usage logs are sent only to our servers, and are stored by us. However, in the web view that the app opens without showing it for the check against misuse (Cloudflare Turnstile; Section 3), Cloudflare may store information on the device in accordance with its own policy (https://www.cloudflare.com/turnstile-privacy-policy/).

Our public pages (https://phrasy.net/privacy / https://phrasy.net/terms / https://phrasy.net/support) do not use cookies, and we do not store browsing records (except for the connection records that the hosting provider keeps; Section 3). The admin screen used only by the operator uses cookies in order to maintain the login state.

The LP (Section 2) records visits as follows, without using cookies or any identifier on the device. We do not use third-party analytics tools. As the visitor mark, we create and record a hash made by mixing the site host, the IP address and the type of browser (User-Agent) into a salt that is valid only for that day. We do not store the IP address or the User-Agent themselves. Because the salt is deleted when the day changes, it is not possible afterwards to find the original person by trying IP addresses, and the same person has a different mark on a different day. We also store the language of the LP, the country (determined from the connection source), the type of device (iOS, Android, PC or other), the referrer's hostname (the path and search terms are discarded), the advertising UTM and campaign identifiers, the fact that the App Store button was pressed and where, and the time. The purpose is to understand, in a form that does not identify individuals, the number of visitors to the LP, where they come from and the rate at which the button is pressed, and to improve advertisements and the LP (Section 4; the basis is the legitimate interests in Section 5). We do not record access when the browser sends the Do Not Track or Global Privacy Control signal, or access from automated crawling programs.

In the iOS app, only to those who allowed it on the explanation screen in the app, we ask for the device's tracking permission (App Tracking Transparency). Only for those who have allowed both, the Meta SDK sends the information in Section 7, including the advertising identifier (IDFA), to Meta Platforms, Inc. This falls under what Apple calls "tracking". The features and prices of the Service do not change if you do not allow it. You can withdraw your permission at any time in "Settings". When you withdraw it, we do not send events or the advertising identifier. The SDK is not run from the next launch. Except where you have given this permission, we do not link users' or devices' information with other companies' apps, websites or offline information in order to deliver advertising or measure its effectiveness (what Apple calls "tracking"). Nor do we provide information to data brokers.

Where we place advertisements, we use mechanisms provided by Apple to measure their effectiveness. For Apple's own advertisements shown on the App Store and elsewhere (Apple Ads), the app obtains from the device, after its first launch, a token for an attribution lookup (Apple's AdServices) and sends it to our server, and we query Apple with it. What we receive from Apple is whether the installation resulted from an Apple advertisement and, if so, the numbers of the campaign, ad group, keyword and ad, the country or region, and whether it was a new download or a redownload. We link this to the record of the user's installation and use it to measure which advertisements led to use of the Service. The token is used only for the lookup and is not stored. This mechanism does not use the advertising identifier (IDFA), and does not link the information with activity in other companies' apps or websites. For advertisements other than Apple's, we use SKAdNetwork. Under that mechanism, the matching of an advertisement click to an app installation is carried out on Apple's side, what we and the advertising provider receive is only aggregated values that cannot identify an individual, and we cannot know which user came from which advertisement. However, for iOS users who have allowed it both on the explanation screen in the app and in the device's tracking permission, the Meta SDK sends information to Meta Platforms, Inc. as described above (Section 7).

Notice of external transmission (Article 27-12 of the Telecommunications Business Act). Where the user's app sends information to a party other than us, from the device or through a function of the device's OS, the recipient, the information sent and the purpose are as follows.

On the Android version (applicable from the time we start providing it), information is likewise passed to Google LLC through Google Play billing, the Play Integrity API, FCM (including the Firebase installation ID) and Google sign-in (Section 7).

16. Notifications

The notifications of the Service are of two kinds: local notifications scheduled on the device (notices about reviews, the notice before the free trial ends, and, on a device that has not registered a notification destination with us, notices about your streak), and notifications that reach you from our server through the Apple Push Notification service (APNs) or, on the Android version (applicable from the time we start providing it), Google's Firebase Cloud Messaging (FCM) (notices when your streak is about to be broken, messages from a Chat partner, and the start, update and end of a Live Activity). Notifications from the server are sent only to devices on which you have allowed notifications. A notification contains, for a streak notice, only the number of days of your streak and fixed wording, and, for a Chat message, the one message the Chat partner wrote (the same message is also stored in Chat); it does not contain recordings. Because notifications from the server go through APNs, the notification destination and the content are passed to Apple (Section 7). On the Android version, notifications from the server go through FCM, so the notification destination and the content are passed to Google (Section 7; Google is our subcontractor).

Notifications can be stopped at any time from the device's settings or from "Profile > Notifications" in the app. A Live Activity can also be stopped in the device's settings (under Phrasy). If you turn notifications off in the app or sign out, we revoke the notification destination and no longer send notifications from the server. The fact that you allowed notifications and that you opened a notification is recorded as a usage log (Section 3).

17. Amendments to this Policy

We may amend this Policy in response to amendments of the law or changes to the Service. When we amend it, we post the amended content and the effective date on this page (https://phrasy.net/privacy).

Where we substantially change the items of information collected, the purposes of use, the subcontractors, or the content of cross-border transfers, we post this on this page at least two weeks before the effective date, and also inform you in the app. Where the change is one that requires new consent, we show the explanation and consent screen again, and apply the change only after obtaining consent. Where consent is not given, the handling up to that point remains unchanged. However, an optional handling that we carry out only where you individually allow it on an explanation screen applies from the time you allow it, after it is posted on this page. The handling for those who do not allow it does not change.

The current version and the effective date are shown at the top of this page. If you need a past version, please request it from the contact point.

18. Contact

For this Policy, the handling of personal information, the exercise of rights, and complaints, please contact the following.

You can also send us defects and requests from "Profile > Send a report" in the app (no contact details need to be entered; if you need a reply, please contact us by email).