Pieces policy evolution
Before/after stance changes across captured policy versions, with exact citations. If no before/after delta is available yet, AIRIN shows the latest citation-backed stance events instead.
No before/after stance delta is available for this filter yet. Latest citation-backed stance events are shown below.
Aug 17, 2026traininglow
model training
Latest stance: no training claim
“This license includes the right to display Your Content and to share Your Content with other users at your request. Except to the extent you have enabled optional analytics or cloud backup, data transmitted to a cloud LLM is processed to return a result and is not retained on our cloud. We do not use Your Content to train, fine-tune, or improve any Pieces AI or machine learning model (see our Privacy Policy for details). You are solely responsible for Your Content and for any harm resulting from it. We are not responsible for the content our users post and share via the Services, and we are not responsible for any misuse of Your Content. We have the right to refuse or remove any of Your Content that, in our sole discretion, violates any law or this Agreement.”
Open citationJul 20, 2026traininglow
model training
Latest stance: no training claim
“This license includes the right to display Your Content and to share Your Content with other users at your request. Except to the extent you have enabled optional analytics or cloud backup, data transmitted to a cloud LLM is processed to return a result and is not retained on our cloud. We do not use Your Content to train, fine-tune, or improve any Pieces AI or machine learning model (see our Privacy Policy for details). You are solely responsible for Your Content and for any harm resulting from it. We are not responsible for the content our users post and share via the Services, and we are not responsible for any misuse of Your Content. We have the right to refuse or remove any of Your Content that, in our sole discretion, violates any law or this Agreement.”
Open citationJul 20, 2026traininghigh
data sharing
Latest stance: sale or sell
“Stored on your device. Your memories and content are saved locally on your device. By default, we do not store your memories or content on our cloud. The limited exceptions are optional cloud backup (if you enable it), usage analytics you have left enabled, and data synced through connectors you set up. Processed with cloud LLMs. To form, enrich, and summarize your memories, and to answer chat prompts, Pieces sends relevant text, visual, and audio-derived data from your device to a cloud LLM — either a Pieces-provisioned default model or your organization's own BYOK model. This processing is transient and is not retained on our cloud afterward. Some processing stays on-device. Capture, text extraction (OCR), retrieval, and search run on your device, and — where your device supports it — audio transcription runs locally too. Sensitive data is redacted before cloud processing. Before content is sent to a cloud LLM, Pieces runs an on-device redaction step that is designed to remove personal identifiers, credentials, keys, and financial data. Automated redaction is not perfect and may not catch everything. You choose what Pieces captures. You control which capture sources (screen, clipboard, audio, files, browser) are enabled, can allow/deny specific apps and websites, and can disable usage analytics. No training on your data. We never use your content, personal information, or Google user data to train, fine-tune, or improve any Pieces AI/ML model. Our proprietary models are trained on synthetic datasets. We do not sell your personal information and do not use it for cross-context behavioral advertising.”
Open citationJul 20, 2026trainingmedium
data sharing
Latest stance: third party or vendor sharing
“We never use your data to train our models. Your content, personal information, and Google user data are never used to train, fine-tune, evaluate, or improve any Pieces AI or machine learning model, whether generalized or personalized. This applies to all product components, including Long-Term Memory, Copilot interactions, context injection, transcription, and metadata generation. Our proprietary on-device models are trained exclusively on synthetic datasets generated using non-user-derived methods. When Pieces sends your data to a cloud LLM to form, enrich, or summarize memories or to generate a chat response, the model is invoked solely to produce the requested output for you in that moment. Cloud LLM providers. When Pieces uses a default cloud model, the model is operated by a third-party LLM provider (reached through our LLM routing provider) acting as our subprocessor under terms that do not permit using your data to train that provider's models. When your organization uses a BYOK model, your data is processed by the provider your organization has configured, subject to that provider's and your organization's terms rather than Pieces's default arrangements. We may use de-identified or aggregated data for legitimate business purposes such as abuse detection, cost verification, and service improvement, and we maintain such data in de-identified form. Where you have left usage analytics enabled, we may also use that data to improve the Services; you can disable analytics at any time in the application's privacy settings.”
Open citationJul 20, 2026traininghigh
data sharing
Latest stance: sale or sell
“We do not use Google user data for any purpose other than providing or improving these user-facing features. Affirmative prohibitions. Pieces does not, and will not: Use Google user data to serve advertisements of any kind, including retargeted, personalized, or interest-based advertisements. Transfer or sell Google user data to advertising platforms, data brokers, or information resellers. Use Google user data for any general marketing, promotional, or business-development purpose. Use Google user data to develop, train, fine-tune, evaluate, or improve any artificial intelligence or machine learning model, whether generalized, personalized (including per-user models), or operated by a third party. Where Pieces uses AI/ML at inference time to deliver a feature you have specifically requested on your Google user data (such as summarization of content you have explicitly directed Pieces to process), the model is invoked solely to produce the requested output for you in that moment; your Google user data is not retained, logged, or otherwise used to train, fine-tune, evaluate, or develop any model, and is not used to improve any model serving you, other users, or any third party. Allow humans to read Google user data, except: (i) with your affirmative, in-product consent for specific Google user data, requested at the point of access; (ii) as necessary for security purposes (such as investigating abuse or a security incident); (iii) to comply with applicable law; or (iv) where the data has been aggregated and anonymized for limited internal operations such as service reliability, abuse prevention, security, and legal compliance, in accordance with the Limited Use requirements.”
Open citationJul 20, 2026traininglow
model training
Latest stance: no training claim
“Mesh Intelligent Technologies, Inc.'s Pieces ("Pieces", "we", "us", or "our") knows that you care about how your personal information is used and shared, and we take your privacy seriously. In this Privacy Policy, we describe how we collect, use, and handle your personal information when you use our services, software, technology, and websites ("Services"). This Privacy Policy applies when Pieces is the data controller with regard to the personal information referenced in it. This Privacy Policy does not apply where Pieces acts as a data processor on behalf of an organization — for example, where your employer provisions Pieces for you under an enterprise agreement. In that case, that organization's agreement (including any Data Processing Addendum) governs, and you should direct privacy requests to your organization. Pieces stores your data locally on your device . By default, we do not store your memories or content on our cloud. To make your memories useful, however, Pieces uses large language models ("LLMs") running in the cloud: as part of forming, enriching, and summarizing your memories, and when generating chat responses, Pieces sends relevant textual, visual, and audio-derived data from your device to a cloud LLM to be processed. That cloud LLM is either a default model provisioned by Pieces or, for members of an organization, a model provisioned by their organization and connected via "Bring Your Own Key" (BYOK). This cloud processing is transient — the data is used to return a result and is not retained on our cloud afterward. We never use your content to train our models. This Privacy Policy explains these practices in detail. Pieces's address is: Mesh Intelligent Technologies, Inc. 1311 Vine St., Unit 301”
Open citationJul 20, 2026traininglow
model training
Latest stance: no training claim
“Stored on your device. Your memories and content are saved locally on your device. By default, we do not store your memories or content on our cloud. The limited exceptions are optional cloud backup (if you enable it), usage analytics you have left enabled, and data synced through connectors you set up. Processed with cloud LLMs. To form, enrich, and summarize your memories, and to answer chat prompts, Pieces sends relevant text, visual, and audio-derived data from your device to a cloud LLM — either a Pieces-provisioned default model or your organization's own BYOK model. This processing is transient and is not retained on our cloud afterward. Some processing stays on-device. Capture, text extraction (OCR), retrieval, and search run on your device, and — where your device supports it — audio transcription runs locally too. Sensitive data is redacted before cloud processing. Before content is sent to a cloud LLM, Pieces runs an on-device redaction step that is designed to remove personal identifiers, credentials, keys, and financial data. Automated redaction is not perfect and may not catch everything. You choose what Pieces captures. You control which capture sources (screen, clipboard, audio, files, browser) are enabled, can allow/deny specific apps and websites, and can disable usage analytics. No training on your data. We never use your content, personal information, or Google user data to train, fine-tune, or improve any Pieces AI/ML model. Our proprietary models are trained on synthetic datasets. We do not sell your personal information and do not use it for cross-context behavioral advertising.”
Open citationJul 20, 2026traininglow
model training
Latest stance: no training claim
“We never use your data to train our models. Your content, personal information, and Google user data are never used to train, fine-tune, evaluate, or improve any Pieces AI or machine learning model, whether generalized or personalized. This applies to all product components, including Long-Term Memory, Copilot interactions, context injection, transcription, and metadata generation. Our proprietary on-device models are trained exclusively on synthetic datasets generated using non-user-derived methods. When Pieces sends your data to a cloud LLM to form, enrich, or summarize memories or to generate a chat response, the model is invoked solely to produce the requested output for you in that moment. Cloud LLM providers. When Pieces uses a default cloud model, the model is operated by a third-party LLM provider (reached through our LLM routing provider) acting as our subprocessor under terms that do not permit using your data to train that provider's models. When your organization uses a BYOK model, your data is processed by the provider your organization has configured, subject to that provider's and your organization's terms rather than Pieces's default arrangements. We may use de-identified or aggregated data for legitimate business purposes such as abuse detection, cost verification, and service improvement, and we maintain such data in de-identified form. Where you have left usage analytics enabled, we may also use that data to improve the Services; you can disable analytics at any time in the application's privacy settings.”
Open citationJul 20, 2026traininglow
model training
Latest stance: no training claim
“We do not use Google user data for any purpose other than providing or improving these user-facing features. Affirmative prohibitions. Pieces does not, and will not: Use Google user data to serve advertisements of any kind, including retargeted, personalized, or interest-based advertisements. Transfer or sell Google user data to advertising platforms, data brokers, or information resellers. Use Google user data for any general marketing, promotional, or business-development purpose. Use Google user data to develop, train, fine-tune, evaluate, or improve any artificial intelligence or machine learning model, whether generalized, personalized (including per-user models), or operated by a third party. Where Pieces uses AI/ML at inference time to deliver a feature you have specifically requested on your Google user data (such as summarization of content you have explicitly directed Pieces to process), the model is invoked solely to produce the requested output for you in that moment; your Google user data is not retained, logged, or otherwise used to train, fine-tune, evaluate, or develop any model, and is not used to improve any model serving you, other users, or any third party. Allow humans to read Google user data, except: (i) with your affirmative, in-product consent for specific Google user data, requested at the point of access; (ii) as necessary for security purposes (such as investigating abuse or a security incident); (iii) to comply with applicable law; or (iv) where the data has been aggregated and anonymized for limited internal operations such as service reliability, abuse prevention, security, and legal compliance, in accordance with the Limited Use requirements.”
Open citationJun 17, 2026traininglow
model training
Latest stance: no training claim
“Use Google user data to develop, train, fine-tune, evaluate, or improve any artificial intelligence or machine learning model, whether generalized, personalized (including per-user models), or operated by a third party. Where Pieces uses AI/ML at inference time to deliver a feature you have specifically requested on your Google user data (such as summarization of content you have explicitly directed Pieces to process), the model is invoked solely to produce the requested output for you in that moment; your Google user data is not retained, logged, or otherwise used to train, fine-tune, evaluate, or develop any model, and is not used to improve any model serving you, other users, or any third party.”
Open citationGenerated from live stance events. Informational only, not legal advice.