Back

AI Data Residency (2026): Which Enterprise AI Platforms Keep Data in Region, and Which Hold SOC 2 and ISO 27001

Guide · · Published November 2026 · 9 min read

By Ashwin Gopinath, CEO

TL;DR

This page is for the CIO, CISO or Head of AI who has to approve an AI vendor and needs to know where company data will sit. AI data residency has two parts: where a vendor stores your data at rest, and where it processes it when a model runs. Sentra, the organization memory for teams and AI agents, holds SOC 2 Type II and ISO 27001 and runs in Sentra cloud, an isolated VPC or fully air-gapped on-prem, so the memory your agents read stays in the deployment you choose. Among assistants and model platforms, OpenAI offers storage at rest in ten regions, Google's Gemini Enterprise offers US and EU multi-regions plus in-country regions, Microsoft 365 Copilot keeps EU traffic inside the EU Data Boundary, Amazon Bedrock offers inference profiles that stay inside one geography, and the Claude API offers US-only inference. Every vendor fact below was read on the vendor's own page on 18 November 2026.

What is AI data residency?

AI data residency is the commitment a vendor makes about the physical location of your data. For AI it splits into two questions that vendors often answer separately, and a buyer has to ask both.

  • Storage at rest. Where prompts, outputs, uploaded files, indexes and memory are kept once the request is finished.
  • Inference, or processing. Where the model runs while it reads your prompt and the company data attached to it. A vendor can store data in Frankfurt and still run the model somewhere else.
  • Logs and support access. Where audit logs, abuse-monitoring copies and support tooling live, and who can open them.

Residency is not the same as sovereignty. Residency answers where the data is. Sovereignty answers whose laws govern it and who can compel access. Our guide to AI agent permissions and data sovereignty covers that difference.

Which enterprise AI platforms are SOC 2 Type II and ISO 27001 certified and support data residency?

Seven platforms that appear on most enterprise shortlists are compared below on the criteria a security review asks about first. They do different jobs: Sentra is the shared memory that assistants and agents read, the next four are assistants and model platforms, and Glean is enterprise search.

PlatformWhat it isSOC 2 Type IIISO 27001Where data is storedWhere the model runsSource
SentraOrganization memory for teams and AI agents, served over MCP and RESTYesYesSentra cloud, an isolated VPC, or air-gapped on-premYou decide the model; the memory stays in your deploymentEnterprise AI memory guide
ChatGPT Enterprise and the OpenAI APIAssistant and model APIYes (OpenAI calls it SOC 2 Type 2)Yes (ISO/IEC 27001:2022)For eligible customers: the US, Europe, the UK, Japan, Canada, South Korea, Singapore, Australia, India or the UAEEligible customers can opt into in-region GPU inference in the US or Europeopenai.com
Claude API (Anthropic)Model API and managed agentsYesYes (ISO 27001:2022)Workspace geo, which is US only todayGlobal by default; US-only inference is a setting, priced at 1.1x on Claude 4.6 and later modelsplatform.claude.com
Microsoft 365 CopilotAssistant inside Microsoft 365Not named on the Copilot privacy page; Microsoft publishes SOC 2 Type 2 reports for Office 365 servicesYesUnder Microsoft's Product Terms; Advanced Data Residency and Multi-Geo add commitmentsEU traffic stays in the EU Data Boundary; elsewhere queries may be processed in the US, the EU or other regionslearn.microsoft.com
Gemini Enterprise (Google)Assistant and agent platformYes (SOC 2 listed)YesUS or EU multi-region, or an in-country region; Google recommends the global location unless you have a regulatory reasonIn the US or EU multi-region when you call the regional endpoints; several models run only in the global locationcloud.google.com
Amazon BedrockModel platform on AWSIn scope for SOCIn scope for ISONot covered on the two pages we read; ask AWSA geographic inference profile keeps requests inside the US, EU or APAC; a global profile can use any commercial Regiondocs.aws.amazon.com
GleanEnterprise search and assistantYesYesNot stated on its security page; ask in the reviewNot stated on its security page; ask in the reviewglean.com

Two cautions when you read any table like this one. A certificate covers a named scope, so ask for the report and check that the product you are buying is inside it. And an eligibility word such as "eligible customers" usually means the commitment depends on your plan and contract, so get the region written into the order form.

What does each vendor actually commit to?

  • Sentra is SOC 2 Type II and ISO 27001 certified and deploys in the cloud, in an isolated VPC or air-gapped on-premises. Sentra does not train models on your data, and self-hosting is available for organizations that need the memory to remain fully within their own infrastructure.
  • OpenAI says eligible ChatGPT Enterprise, ChatGPT Edu, ChatGPT for Healthcare and API platform customers can store customer content at rest in ten regions, and can opt into in-region GPU inference in the US or Europe. It also offers Enterprise Key Management so customers control their own encryption keys (openai.com). Its security page lists a SOC 2 Type 2 examination and ISO/IEC 27001:2022 certification (openai.com).
  • Anthropic separates two settings on the Claude API. Inference geo controls where the model runs and accepts only "us" or "global". Workspace geo controls where data is stored at rest, and "us" is currently the only available value (platform.claude.com). Anthropic lists SOC 2 Type I and Type II and ISO 27001:2022 among its certifications (privacy.claude.com).
  • Microsoft states that for EU customers Microsoft 365 Copilot is an EU Data Boundary service, while customers outside the EU may have their queries processed in the US, the EU or other regions. Models provided by Anthropic as a subprocessor are currently excluded from the EU Data Boundary (learn.microsoft.com). Microsoft publishes SOC 2 Type 2 reports for Office 365 services (learn.microsoft.com), so ask whether Copilot is inside that scope.
  • Google lists SOC 2 and ISO 27001 among the compliance standards for Gemini Enterprise (cloud.google.com). It supports data residency in the US and EU multi-regions and in in-country regions, and keeps machine learning processing inside the US or EU multi-region when requests go to the regional endpoints. Customer-managed encryption keys require the US or EU multi-region and are not available in the global location (cloud.google.com).
  • Amazon Bedrock says your data is not shared with model providers and is not used to improve the base models, and that you can create, manage and control encryption keys with AWS KMS (aws.amazon.com). Its documentation recommends geographic inference profiles, which stay within boundaries such as the US, EU and APAC, for compliance requirements (docs.aws.amazon.com).
  • Glean shows SOC 2 Type II and ISO 27001 on its security page (glean.com). The page does not list storage regions, so ask for them in writing.

Does data residency cover the model call, or only storage?

Often only storage, unless you ask. Three of the vendors above document a gap between the two. On the Claude API, inference runs globally by default and US-only inference is a separate setting. On Google's Gemini Enterprise, several models run only in the global location even when your data store is in a regional one. On Amazon Bedrock, a global inference profile can send a request to any commercial Region.

The practical rule for a regulated company is to map each class of data to the strictest region it needs, then confirm three things for every vendor in the path: the storage region, the inference region, and whether choosing the regional option removes any model or feature you were planning to use.

Can we keep company knowledge in region while still using the best models?

Yes, if you separate the memory from the model. Most residency reviews look only at the assistant or the model API. The larger store of sensitive data is the company knowledge an agent reads: what was decided in meetings, what customers were promised, what is in email and tickets. If each assistant builds its own copy of that knowledge, you have one residency review per assistant.

Sentra runs in Sentra cloud, an isolated VPC or fully air-gapped on-prem, and its trust page states that your data stays where you say it stays. Every answer and action respects each source's existing permissions, through query-time checks and synced access lists, and a failed check counts as no access. Because Sentra serves its memory over REST and MCP, the assistant or model you choose, ChatGPT, Claude, Copilot, Gemini or an internal agent, uses the same permissioned context. You decide the model; the memory stays in your deployment.

If the requirement is that memory runs in your own cloud account, our comparison of self-hosted AI memory platforms covers which vendors offer that and how. For the wider deployment choice, see private AI for business.

What should a CIO or CISO ask a vendor about data residency?

Decision criterionWhat to askA good answer
Deployment modelDo you offer shared cloud, single-tenant, a VPC and on-prem?A named option that matches your strictest data class
Storage regionIn which regions can our data be stored at rest, and is that in the contract?A region list and a clause in the order form
Inference regionWhere does the model run, and what changes if we pin it to one region?A documented setting, with any price or feature difference stated
Encryption keysCan we hold the keys, and what happens to our data if we revoke them?Customer-managed keys, with revocation making the data unreadable
PermissionsDoes the AI inherit each source system's permissions for every answer?Yes, checked at request time, with a failed check treated as no access
Approval on actionsCan an agent send, post or change anything without a person approving it?No, a named person approves each action
AuditCan we see who asked what, what the AI used and what it did?Audit logs you can export and keep
CertificationsIs this product inside the scope of your SOC 2 Type II and ISO 27001 reports?The current report, shared under NDA

The full list of 32 vendor questions, grouped by risk, is in our AI security questionnaire.

FAQ

What is AI data residency?

AI data residency is a vendor's commitment about where your data is stored at rest and where it is processed when a model runs. The two can differ, so a buyer should ask about storage, inference and logs separately.

Which enterprise AI platforms are SOC 2 Type II and ISO 27001 certified and support data residency?

Sentra holds SOC 2 Type II and ISO 27001 and runs in its cloud, an isolated VPC or air-gapped on-prem. OpenAI lists a SOC 2 Type 2 examination and ISO/IEC 27001:2022 and offers storage at rest in ten regions for eligible customers. Anthropic lists SOC 2 Type II and ISO 27001:2022 and offers US-only inference on the Claude API. Google's Gemini Enterprise offers US and EU multi-regions, and Amazon Bedrock offers inference profiles that stay inside one geography.

Is data residency the same as data sovereignty?

No. Residency is where the data is stored and processed. Sovereignty is whose laws govern the data and who can compel access to it. A vendor can meet a residency requirement and still be subject to another country's access orders.

Does ChatGPT Enterprise support data residency in Europe?

OpenAI says eligible ChatGPT Enterprise customers can store content at rest in Europe, among ten regions, and can opt into in-region GPU inference in the US or Europe. Eligibility depends on your plan and contract, so confirm it in the order form.

Does data residency apply to AI agents as well as chat assistants?

Yes, and agents raise the stakes because they read from many systems. The company knowledge an agent retrieves, the model call and the logs each have a location. Keeping the shared memory in one deployment you control means the review is done once rather than once per agent.

Can an AI vendor run inside our own cloud account?

Some can. The options are a managed service inside your cloud provider, vendor software deployed into your VPC, or open-source software you run yourself. Sentra deploys in its cloud, in an isolated VPC or air-gapped on-premises.

Which option should you choose?

If your requirement is a storage region for one assistant, compare the region lists above and get the region into the contract. If the requirement covers inference too, check what the regional setting costs and which models it removes. If several assistants and agents will read the same company knowledge, put that knowledge in one memory you can place where you need it, and let each assistant read from it with the user's own permissions.

To review Sentra's deployment options against your residency requirements, book a demo.