What Amazon Bedrock, Microsoft Foundry and Gemini Enterprise Agent Platform say on their own pages happens to your prompts and outputs when you call a model there.
Words in the grid: Not stated
What this page covers, and what it does not
6 combinations of model, tier and region recorded
Combinations recorded
| Question | Answer |
|---|---|
| Is this combination offered? | Not stated. |
| Can the model's maker see prompts and outputs? | Model providers cannot view Amazon Bedrock logs or customer prompts and completions because they lack access to the Model Deployment Accounts. AWS, Data protection, read 2026-10-21 Where on the pageNear the top of the page |
| Who processes the data? | When using Amazon Bedrock or Google Cloud's Agent Platform, the respective cloud provider acts as the data processor, and users should consult those platforms' compliance and retention materials for matching controls. Anthropic, API and data retention - Claude Platform Docs, read 2026-09-24 Where on the pageNear the top of the page |
| Where are prompts and outputs processed? | Applies to A, C Data processing stays strictly inside one Region. AWS, Regional availability by models, read 2026-09-24 Where on the pageInference options at a glance, in the table "Inference options at a glance", row "Data residency", column "In-Region" Applies to B Prompts and outputs can move within geographic boundaries, such as all EU Regions, but cannot leave that geography. AWS, Regional availability by models, read 2026-09-24 Where on the pageInference options at a glance, in the table "Inference options at a glance", row "Data residency", column "Geographic (Geo) Cross-Region" Applies to D, F The same passage as the answer for A, C. More general passageWhen using Amazon Bedrock or Google Cloud's Agent Platform, stored information remains inside the respective cloud provider's infrastructure, and users should check each platform's guides for setup instructions. Anthropic, API and data retention - Claude Platform Docs, read 2026-10-21 Where on the pageModel-specific data retention requirements Applies to E The same passage as the answer for B. More general passageThe same passage as the answer for D, F. |
| Where is data at rest stored? | Applies to A, B, D, E Not stated. Applies to C, F Data in AWS GovCloud (US) stays in those Regions, except certain data components that the page lists as able to leave them in normal operation. AWS, Amazon Bedrock in AWS GovCloud (US), read 2026-10-21 Where on the pageExport-controlled content |
| What retention options are there? | Applies to A, B, C For models like Claude Opus 4.8 that allow the "none" mode, data is never retained, reviewed, or shared, regardless of the account or project settings. AWS, Data retention, read 2026-10-21 Where on the pageModel availability and data retention, in a list, in a highlighted note Applies to D, E, F Claude Fable 5 is an example of a model whose only permitted settings are AWS's human review and a provider data-sharing setting; such a model requires human review, so its data is kept inside the AWS boundary, AWS may review it on any request, and the model provider receives none of it. AWS, Data retention, read 2026-10-21 Where on the pageModel availability and data retention, in a list, in a highlighted note |
| How long is data kept? | Applies to A, B, C The same passage as the answer under “What retention options are there?” Applies to D, E, F For models that require AWS's own human review, currently Claude Fable 5 and Claude Fable 5.1, prompts and completions are kept inside the AWS boundary for 30 days at most and AWS may review them to meet the model provider's access condition; the content does not go to the model provider. AWS, Data retention, read 2026-10-21 Where on the pageWhat data is retained and for how long More general passageThe obligation to keep data for 30 days exists in all locations where Covered Models are made available. Anthropic, API and data retention - Claude Platform Docs, read 2026-10-21 Where on the pageModel-specific data retention requirements |
| Do people review the data? | Applies to A, B, C The same passage as the answer under “What retention options are there?” More general passageAWS's Bedrock documentation: when a model does not need human review, the company will not examine your material. AWS, Data retention, read 2026-10-21 Where on the pageData retention modes, in the table "Data retention modes", column "Behavior" Applies to D, E, F The same passage as the answer under “How long is data kept?” More general passageThe same passage as the answer for A, B, C. |
| Is the data used for training? | Not stated. |
| Is the data shared with third parties? | Applies to A, B, C The same passage as the answer under “What retention options are there?” Applies to D, E, F The same passage as the answer under “How long is data kept?” |
The same passage as the answer under “Where are prompts and outputs processed?”
The same passage as the answer under “Where are prompts and outputs processed?”
There are no geographic limits, allowing data to be handled in any commercial Region.
Inference options at a glance, in the table "Inference options at a glance", row "Data residency", column "Global Cross-Region"
Pages read for this cloud: Route model inference requests across AWS Regions with cross-Region inference, read 2026-09-24; Data protection, read 2026-10-21; Data retention, read 2026-10-21; Regional availability by models, read 2026-09-24; Amazon Bedrock in AWS GovCloud (US), read 2026-10-21; API and data retention - Claude Platform Docs, read 2026-09-24. Unverified after 2026-12-23.
8 combinations of model, tier and region recorded
Combinations recorded
| Question | Answer |
|---|---|
| Is this combination offered? | Applies to A, B, C, G, H Not stated. Applies to D Microsoft's table marks this combination "✅". Microsoft, Foundry Models sold by Azure in Azure Government, read 2026-10-21 Where on the pageModel summary table and region availability › Models by deployment type › Data Zone Standard model availability, in the table "Data Zone Standard model availability", row "usgovarizona" More general passageEvery Azure OpenAI model offered in Azure Government counts among the Models sold by Azure. Microsoft, Foundry Models sold by Azure in Azure Government, read 2026-10-21 Where on the pageNear the top of the page Applies to E, F Not listed: Foundry Models sold by Azure in Azure Government, read 2026-09-24, was read for this combination and lists it nowhere. |
| Can the model's maker see prompts and outputs? | Applies to A, B Anthropic staff examine customer content only when necessary to check for possible safety breaches, following their own terms. Microsoft, Data, privacy, and security for Claude models in Microsoft Foundry, read 2026-10-21 Where on the pageHosted on Azure Applies to C A customer's prompts, completions, embeddings and training data are not made available to OpenAI or to any other provider of Models sold by Azure. Microsoft, Data, privacy, and security for Models sold by Azure in Microsoft Foundry, read 2026-10-21 Where on the pageNear the top of the page, in a list, in a highlighted note Applies to D, E, F Not stated. Applies to G, H When using the Hosted on Anthropic Infrastructure option, prompts and outputs are handled on Anthropic's servers, which may involve processing data outside of Azure and outside of the chosen Azure region. Microsoft, Data, privacy, and security for Claude models in Microsoft Foundry, read 2026-10-21 Where on the pageHosted on Anthropic Infrastructure |
| Who processes the data? | Applies to A, B, G, H Under both hosting options, Hosted on Azure and Hosted on Anthropic infrastructure, Anthropic sells and runs the Claude models offered in Microsoft Foundry and is, for their prompts and outputs, an independent data processor; using them is subject to Anthropic's terms for its Claude models and APIs. Microsoft, Data, privacy, and security for Claude models in Microsoft Foundry, read 2026-10-21 Where on the pageNear the top of the page Applies to C Data is stored and processed by Models sold by Azure to deliver the service and to watch for uses that breach the applicable product terms, and it names the Microsoft Products and Services Data Protection Addendum as governing that processing. Microsoft, Data, privacy, and security for Models sold by Azure in Microsoft Foundry, read 2026-10-21 Where on the pageNear the top of the page Applies to D, E, F Not stated. |
| Where are prompts and outputs processed? | Applies to A, B Selecting the Hosted on Azure option means that request handling, API functions, and GPU computing for prompts and outputs occur on Azure systems. Microsoft, Data, privacy, and security for Claude models in Microsoft Foundry, read 2026-10-21 Where on the pageHosted on Azure 2 more general passagesFor the Global Standard deployment type, data processing can take place in any Azure region. Microsoft, Deployment types for Microsoft Foundry Models, read 2026-10-21 Where on the pageDeployment type comparison, in the table "Deployment type comparison", row "Global Standard", column "Data processing" Global tier models might be handled in any Azure location. Microsoft, Deployment types for Microsoft Foundry Models, read 2026-10-21 Where on the pageNear the top of the page, in a list, in a highlighted note Applies to C Prompts and responses are handled inside the customer's chosen geography, excluding Global or DataZone deployments, though they may move between regions within that geography for operational reasons like performance and capacity management. Microsoft, Data, privacy, and security for Models sold by Azure in Microsoft Foundry, read 2026-10-21 Where on the pageHow does Foundry process data to provide Models sold by Azure? › Generating completions, images or embeddings through inferencing More general passageData is processed within Azure geography. Microsoft, Deployment types for Microsoft Foundry Models, read 2026-10-21 Where on the pageDeployment type comparison, in the table "Deployment type comparison", row "Standard", column "Data processing" Applies to D The Standard tier includes a USGov datazone option that routes traffic inside Azure Government for higher throughput. Microsoft, Foundry Models sold by Azure in Azure Government, read 2026-10-21 Where on the pageModel summary table and region availability › Models by deployment type, in a list Applies to E, F Not stated. Applies to G, H The same passage as the answer under “Can the model's maker see prompts and outputs?” |
| Where is data at rest stored? | Applies to A, B When using the Hosted on Azure option, stored data remains in the chosen Azure geography and processing is limited to the "Global" or "DataZone" options. Microsoft, Data, privacy, and security for Claude models in Microsoft Foundry, read 2026-10-21 Where on the pageHosted on Azure Applies to C For the features of Models sold by Azure that keep data in the service, such as uploads through the Files API or vector store, the Responses API, Assistants API threads and Stored completions, that data is held at rest inside the Foundry resource within the customer's own Azure tenant, in the resource's own geography. Microsoft, Data, privacy, and security for Models sold by Azure in Microsoft Foundry, read 2026-10-21 Where on the pageHow does Foundry process data to provide Models sold by Azure? › Data storage for Models sold by Azure features, in a list Applies to D, E, F, G, H Not stated. |
| What retention options are there? | Applies to A, B, D, E, F, G, H Not stated. Applies to C If a customer is approved for modified abuse monitoring, the abuse-monitoring storage of prompts and completions, and the human review of them, which the page describes just before, are not carried out. Microsoft, Data, privacy, and security for Models sold by Azure in Microsoft Foundry, read 2026-10-21 Where on the pageHow does Foundry process data to provide Models sold by Azure? › Preventing abuse |
| How long is data kept? | Not stated. |
| Do people review the data? | Applies to A, B Automated systems identify material that could be forwarded to Anthropic's Trust & Safety team for evaluation. Microsoft, Data, privacy, and security for Claude models in Microsoft Foundry, read 2026-10-21 Where on the pageHosted on Azure Applies to C Human review is done by authorized Microsoft employees, who reach data through point-wise queries by request ID on Secure Access Workstations with Just-In-Time approval from team managers, and where Models sold by Azure are deployed within the European Economic Area, the reviewers sit in that same area. Microsoft, Data, privacy, and security for Models sold by Azure in Microsoft Foundry, read 2026-10-21 Where on the pageHow does Foundry process data to provide Models sold by Azure? › Preventing abuse Applies to D, E, F, G, H Not stated. |
| Is the data used for training? | Applies to A, B, D, E, F, G, H Not stated. Applies to C A customer's prompts, completions, embeddings and training data do not go into training any generative AI foundation model unless the customer permits or instructs it. Microsoft, Data, privacy, and security for Models sold by Azure in Microsoft Foundry, read 2026-10-21 Where on the pageNear the top of the page, in a list, in a highlighted note |
| Is the data shared with third parties? | Applies to A, B, D, E, F, G, H Not stated. Applies to C Your prompts, completions, embeddings, and training data are not accessible to other customers. Microsoft, Data, privacy, and security for Models sold by Azure in Microsoft Foundry, read 2026-10-21 Where on the pageNear the top of the page, in a list, in a highlighted note Also: Foundry is one of Azure's services, Microsoft hosts Models sold by Azure within its own Azure environment, and those models have no interaction with any service that their providers run, OpenAI for instance, whether the OpenAI API or ChatGPT. Microsoft, Data, privacy, and security for Models sold by Azure in Microsoft Foundry, read 2026-10-21 Where on the pageNear the top of the page, in a highlighted note |
The same passage as the answer under “Where is data at rest stored?”
The same passage as the answer under “Where are prompts and outputs processed?”
The same passage as the answer under “Where are prompts and outputs processed?”
The same passage as the answer under “Where is data at rest stored?”
For the Data Zone Standard deployment type, data processing stays inside the data zone.
Deployment type comparison, in the table "Deployment type comparison", row "Data Zone Standard", column "Data processing"
The same passage as the answer under “Where are prompts and outputs processed?”
Pages read for this cloud: Microsoft Foundry in Azure Government, read 2026-09-24; Deployment types for Microsoft Foundry Models, read 2026-10-21; Foundry Models sold by Azure in Azure Government, read 2026-09-24; Data, privacy, and security for Claude models in Microsoft Foundry, read 2026-10-21; Data, privacy, and security for Models sold by Azure in Microsoft Foundry, read 2026-10-21. Unverified after 2027-01-19.
6 combinations of model, tier and region recorded
Combinations recorded
| Question | Answer |
|---|---|
| Is this combination offered? | Applies to A, B, D, E Multi-region endpoints support every Claude model at version 4.7 or later, with claude-fable-5, claude-opus-4-7 and claude-opus-4-8 as its examples. Google, Gemini Enterprise Agent Platform partner models for MaaS, read 2026-09-24 Where on the pageMulti-region endpoint More general passageThird-party models like Anthropic's Claude and Mistral can be used on the Gemini Enterprise Agent Platform. Google, Gemini Enterprise Agent Platform partner models for MaaS, read 2026-10-21 Where on the pageNear the top of the page Applies to C, F Not stated. More general passage, for context onlyThe same passage as the answer for A, B, D, E. |
| Can the model's maker see prompts and outputs? | Not stated. |
| Who processes the data? | When the Gemini Enterprise API is used, including with partner models, user inputs and outputs are not given to outside parties, and Google handles such information solely per customer directions as detailed in the Cloud Data Processing Addendum. Google, Gemini Enterprise Agent Platform partner models for MaaS, read 2026-10-21 Where on the pageGrant user access to partner models › Partner model regulatory compliance More general passageWhen using Amazon Bedrock or Google Cloud's Agent Platform, the respective cloud provider acts as the data processor, and users should consult those platforms' compliance and retention materials for matching controls. Anthropic, API and data retention - Claude Platform Docs, read 2026-09-24 Where on the pageNear the top of the page |
| Where are prompts and outputs processed? | Applies to A Google's table marks this combination "Supported". Google, Data residency, read 2026-10-21 Where on the pageSupported models › Google Cloud partner model support, in the table "Google Cloud partner model support", row "Anthropic's Anthropic's Claude Opus 4.8 on Google Cloud" More general passageML processing remains inside the chosen geographic area, like the US or EU, when jurisdictional endpoints are used. Google, Data residency, read 2026-10-21 Where on the pageWhere your data lives and is processed, in a list Applies to B Google's table marks this combination "Supported". Google, Data residency, read 2026-10-21 Where on the pageSupported models › Google Cloud partner model support, in the table "Google Cloud partner model support", row "Anthropic's Anthropic's Claude Opus 4.8 on Google Cloud" 2 more general passagesFor ML processing, the multi-region endpoint for the European Union (eu) holds data residency strictly to EU member states, and places outside the European Union's political boundary, the United Kingdom and Switzerland among them, fall outside that endpoint. Google, Data residency, read 2026-10-21 Where on the pageWhere your data lives and is processed, in a list, in a highlighted note The same passage as the answer for A. Applies to C Left blank in the cloud's table: Google's table leaves this cell empty. Google, Data residency, read 2026-10-21 Where on the pageSupported models › Google Cloud partner model support, in the table "Google Cloud partner model support", row "Anthropic's Anthropic's Claude Opus 4.8 on Google Cloud" 2 more general passages, for context onlyML work done at specific regional addresses stays completely inside the wider multi-regional or national legal area linked to that location. Google, Data residency, read 2026-10-21 Where on the pageWhere your data lives and is processed, in a list Requests are delivered from the region you specify when using regional endpoints. Google, Gemini Enterprise Agent Platform partner models for MaaS, read 2026-10-21 Where on the pageNear the top of the page Applies to D Google's table marks this combination "Supported". Google, Data residency, read 2026-10-21 Where on the pageSupported models › Google Cloud partner model support, in the table "Google Cloud partner model support", row "Anthropic's Anthropic's Claude Fable 5 on Google Cloud" 2 more general passagesThe same passage as the answer for A. When using Amazon Bedrock or Google Cloud's Agent Platform, stored information remains inside the respective cloud provider's infrastructure, and users should check each platform's guides for setup instructions. Anthropic, API and data retention - Claude Platform Docs, read 2026-10-21 Where on the pageModel-specific data retention requirements Applies to E Google's table marks this combination "Supported". Google, Data residency, read 2026-10-21 Where on the pageSupported models › Google Cloud partner model support, in the table "Google Cloud partner model support", row "Anthropic's Anthropic's Claude Fable 5 on Google Cloud" 3 more general passagesThe same passage as the answer for B. The same passage as the answer for A. The same passage as the answer for D. Applies to F Left blank in the cloud's table: Google's table leaves this cell empty. Google, Data residency, read 2026-10-21 Where on the pageSupported models › Google Cloud partner model support, in the table "Google Cloud partner model support", row "Anthropic's Anthropic's Claude Fable 5 on Google Cloud" 3 more general passages, for context onlyThe same passage as the answer for C. The same passage as the answer for C. The same passage as the answer for D. |
| Where is data at rest stored? | Partner models on that platform keep data at rest in the chosen region or multi-region, yet how data processing is regionalized might differ. Google, Gemini Enterprise Agent Platform partner models for MaaS, read 2026-10-21 Where on the pageGrant user access to partner models › Partner model regulatory compliance More general passageData kept at rest in the location chosen by the customer stays in that location, regardless of which Agent Platform endpoint the customer's request uses. Google, Data residency, read 2026-10-21 Where on the pageNear the top of the page |
| What retention options are there? | Not stated. |
| How long is data kept? | Applies to A, B, C Not stated. Applies to D, E, F The obligation to keep data for 30 days exists in all locations where Covered Models are made available. Anthropic, API and data retention - Claude Platform Docs, read 2026-10-21 Where on the pageModel-specific data retention requirements |
| Do people review the data? | Not stated. |
| Is the data used for training? | Citing the Training Restriction in its Service Specific Terms, Google does not fine-tune or train any AI/ML model on customer data without the customer's prior permission or instruction, and this covers every managed model on Gemini Enterprise Agent Platform, GA and pre-GA alike. Google, Gemini Enterprise Agent Platform and zero data retention, read 2026-10-21 Where on the pageTraining restriction |
| Is the data shared with third parties? | The same passage as the answer under “Who processes the data?” |
The same passage as the answer under “Where are prompts and outputs processed?”
The same passage as the answer under “Where are prompts and outputs processed?”
Using global endpoints allows Google to handle requests from any region supported by the model, which may cause increased latency.
Near the top of the page
Pages read for this cloud: Gemini Enterprise Agent Platform partner models for MaaS, read 2026-09-24; Data residency, read 2026-10-21; Gemini Enterprise Agent Platform and zero data retention, read 2026-10-21; API and data retention - Claude Platform Docs, read 2026-09-24. Unverified after 2026-12-23.
The three columns are our grouping, and no cloud uses them. Each cell is our short name for the cloud's tier, and sharing a column does not mean two clouds' tiers reach equally far.
| Cloud | regional | geographic or DataZone | global |
|---|---|---|---|
| AWS (Amazon Bedrock) | In-Region | Geographic | Global |
| Microsoft (Microsoft Foundry) | Azure geography | Data Zone | Global |
| Google (Gemini Enterprise Agent Platform) | Locational endpoints | Jurisdictional multi-region endpoints | Global endpoints |
About this page
Not stated: No sentence on the pages we read for that cloud answers the question for that combination. It says nothing about what the cloud does, or about any contract.
Left blank in the cloud's table: The combination's cell is empty, which we read as neither answer.
Not listed: The cloud's own table, or a page read for it, does not list the combination at all.
Each answer is our summary of a passage on the cloud's own page. The line under it names the company, the page and the date we read it, and "Where on the page" opens to the section. Someone other than the summary's writer checked it against the passage. The page may have changed since, so read it there before a decision turns on it.
Applies to names, by letter, the combinations listed above the cloud's table that an answer is recorded for, where that is not all of them; an answer with no such line applies to all of them. An answer can be bound to a combination it does not name, and applies to a region only as far as it says.
Region beside a combination is the one the combination was recorded for. Where no answer names a region, none of them is specific to it.
More general passages are sentences that cover the combination less narrowly than the answer shown. Where the answer excludes the combination or is blank, they are context and not the answer.
Unverified after is 90 days after the earliest read among a cloud's answers.
Rendered 2026-11-01
Some lines here restate, in the catalog's own words, pages licensed under Creative Commons Attribution 4.0 International: Google Cloud documentation (license stated on this docs.cloud.google.com page, read 2026-09-24) and MicrosoftDocs/azure-ai-docs (license stated on this github.com page, read 2026-09-25). Some lines here restate, in the catalog's own words, pages licensed under Creative Commons Attribution-ShareAlike 4.0 International: docs.aws.amazon.com (license stated on this aws.amazon.com page, read 2026-09-26). Our lines that summarize those pages are changed from the originals and are offered under Creative Commons Attribution-ShareAlike 4.0 International, not under the catalog's own license. Each license applies as its page states it. Our lines are summaries, not the publishers' text, and the publishers do not endorse them.
Our reading of vendors' pages, as of the dates shown. Confirm on the vendor's page before relying on any line that cites it. To report an error write to report@crinaro.ai