AI chatbot subprocessors: who OpenAI, Anthropic, Google, and others pass your data to
The Morse teamPublished 8 min read
As of November 2026, eight AI chatbot companies publish a list of their subprocessors, the outside companies that process data for them: OpenAI, Anthropic, Google, Microsoft, Perplexity, Mistral, xAI, and Cohere. Most of those lists cover only business and API customers, so someone using ChatGPT, Gemini, Copilot, or Grok on a personal account can't find out by name who handles their chats. Anthropic and Mistral are the exceptions, with lists that name their consumer apps.
We read each company's published list on November 8, 2026, counted the entries, recorded the countries, and checked which products each one covers, then read the consumer privacy policies that sit beside them. DeepSeek and Meta, which publish no list, are in the table too. Each of the eight deserves credit for publishing its list without a login, and Google for keeping every old version of its Cloud list since 2017.
What is a subprocessor?
A subprocessor is a company another company hires to handle personal data on its behalf, such as a cloud host, a support desk, or a content moderation firm. Anthropic's data processing addendum defines one as "an entity engaged by Anthropic to process Customer Personal Data" (Anthropic). The word comes from European data protection law. In those contracts the business customer is the controller, the AI company is its processor, and anyone the AI company hires in turn is a subprocessor. That's why the lists are written for business customers. A consumer has no such contract with the AI company, so its privacy policy describes "service providers" instead.
Which AI chatbots publish a subprocessor list?
Eight of the ten companies we checked. This table is built from each list as it read on November 8, 2026. "Affiliates" are the company's own subsidiaries, which every list counts separately.
| Company | Date on the list | Outside companies | Affiliates | Locations | Covers the consumer chatbot? |
|---|---|---|---|---|---|
| OpenAI (list) | October 14, 2026 | 17 rows, 16 companies | 5 | 29 countries, plus Cloudflare's "closest" data center | No, except web pages made with ChatGPT Sites on Plus and Pro |
| Anthropic (list) | None; last change posted October 14, 2026 | 22 | 11, listed in its trust center FAQ | US for 14, "Worldwide" for 5, and one each in Canada, South Africa, and the UK | Yes: rows are tagged "Claude Free/Pro/Max" or "All Products" |
| Google Workspace, incl. the Gemini app for work (list) | October 19, 2026 | 14, plus 49 of their affiliates | 45, in 28 countries | 18 countries | No list for personal Gemini |
| Google Cloud, incl. Gemini on Vertex AI (list) | October 19, 2026 | 37 | 75, in 40 countries | More than 30 countries | Business only |
| Microsoft Online Services (list) | September 10, 2026 | 51 rows, 50 companies | 37 datacenter entities, in 35 countries or regions | Mostly US for contract staff; varies by vendor | No: consumer Copilot falls under the general Privacy Statement |
| Perplexity (list) | None | 23 | None listed | All US | Not stated |
| Mistral (list) | None | 24 | 30, on a separate page | EEA countries, the US, and "Worldwide" | Yes: rows are tagged "Vibe," the chat app formerly called Le Chat |
| xAI, now SpaceXSI (list) | None | 22 | 2 rows | 23 rows US, 1 UK | No: tied to its API data processing agreement |
| Cohere (list) | None | 12 | None listed | All USA | Platform and API users |
| DeepSeek | No list | Stores data "in People's Republic of China" (DeepSeek) | |||
| Meta AI | No list |
A country on a list is where a vendor may process data, not where your history is kept; where AI chat data is stored covers that. Counts also depend on what you count. Google's Workspace list is 14 companies on its face and 63 once their own affiliates are added, and OpenAI lists Cloudflare twice, once for its content delivery network and once for web hosting.
Does the ChatGPT subprocessor list cover your ChatGPT chats?
No, apart from one feature. OpenAI's list opens: "OpenAI engages the following entities to provide processing activities for Customer Data (as defined in the OpenAI Data Processing Agreement)" (OpenAI). Every row is tagged with some mix of API, ChatGPT Enterprise, ChatGPT Edu, and ChatGPT Business. None is tagged Free, Go, Plus, or Pro. The only paragraph for individual plans is headed "Processing for ChatGPT Pro and Plus users," and it covers web pages published with ChatGPT Sites, not chats.
On a personal plan, OpenAI's privacy policy describes its vendors by category and names none, which we go through in is ChatGPT private. A business customer gets names, countries, and 30 days to object to a new one.
Google and Microsoft draw the line the same way. Google's Workspace list covers its "Core Services," which its services summary says include the "Gemini app" for work accounts (Google). We found no list for personal Gemini, whose Privacy Hub says chats reviewed by "Google's trained service providers" are kept for up to three years without naming those providers (Google). Microsoft's list covers services "governed by the Microsoft Data Protection Addendum" (Microsoft), and its Privacy Statement tells consumer Copilot users only that it shares data with "vendors or agents working on our behalf" (Microsoft). xAI's list rests on a data processing agreement for "our API or other business services" (xAI), and its privacy policy for Grok says it "does not apply" to that business data (xAI).
Who are Anthropic's subprocessors?
Twenty-two companies, and they include the ones that check who you are. Anthropic's list has no scoping sentence, and its rows carry product tags instead. Two identity checkers, Persona and Yoti, are tagged "Claude Free/Pro/Max" for "Fraud and abuse detection, identity verification," and Stripe handles billing for "Claude Pro/Max" (Anthropic). The rest:
- Cloud and network: Google Cloud Platform, Amazon Web Services, and Microsoft Azure, all "Worldwide," and Cloudflare.
- Support: Intercom, plus Nutun in South Africa and Boldr in Canada.
- Fraud and abuse: Sift, Arkose Labs, and hCaptcha, alongside Persona and Yoti.
- Web search: Brave Search, TurboPuffer, and Parallel Web Systems, the last added on October 14, 2026 (Anthropic).
- Other: Twilio and Iterable for email and texts, Sentry for error handling, WorkOS for single sign-on, ElevenLabs for text to speech, and Palantir Federal Cloud Service for Claude for Government.
No human review or data labeling firm appears. Anthropic's business customers get 15 days to object to a new subprocessor, half of OpenAI's 30. What Anthropic's own staff can see is in is Claude private.
Do subprocessors read your chats?
Some are hired to. OpenAI names three moderation vendors for business content: TaskUs in the Philippines, Accenture in the US, Canada, and the Philippines, and Cinder, a "Platform for content moderation." Its list says that for content its models flag, "OpenAI may share samples of the flagged Customer Content with relevant Sub-processors," and that they keep samples only "for the period of review" (OpenAI).
Google Cloud names six firms for "Data Labeling," where "Human labelers apply labels to datasets submitted by Customer based on instructions provided by Customer": Cognizant, two GlobalLogic entities, Innodata, Teleperformance, and Uber Technologies, the last in India, Kenya, Poland, and the US (Google). On Workspace, Wipro does "Human review of indicators of account compromise or abuse," with access "limited to suspicious search terms" (Google).
Consumer apps use reviewers too, without naming the firms; who reviews AI chats covers what Google, Microsoft, and others say about them.
Are AI companies each other's subprocessors?
Yes, and the lists show the web of it. Perplexity files four AI labs under "Cloud provider," which we go through in Morse vs Perplexity. Microsoft's list has an "AI subprocessors" category of five: Anthropic, GitHub, Fireworks AI, OpenAI, whose location is given as "Worldwide," and SpaceXAI, added in September 2026 with processing in eight countries, including Saudi Arabia and the United Arab Emirates (Microsoft). OpenAI's own list runs the other way, with Microsoft, Google Cloud, Oracle, Amazon Web Services, CoreWeave, and Cerebras as its compute.
The arrangement can change what a work account does without a new app. Microsoft's page on Anthropic says "Anthropic has onboarded as a Microsoft subprocessor," that Claude models are on "by default for most customers in commercial cloud (excluding EU/EFTA and UK)," and that they're "currently excluded from the EU Data Boundary" (Microsoft). How one government responded is in Sweden blocks ChatGPT and Claude.
How much notice do you get before a new subprocessor?
None by contract, as a consumer. Business customers with a data processing agreement get a set window, from 10 days to six months:
| Company | Notice before a new subprocessor | Source |
|---|---|---|
| Microsoft | "at least 6 months in advance," or "30 days' notice" for AI subprocessors, with the ability to disable them "until at least 6 months after notice" | Microsoft DPA |
| "at least 30 days before the New Subprocessor starts processing," and a customer may object within 90 days by terminating the agreement | Google Cloud DPA | |
| OpenAI | Notice by blog post, in-product, or email; 30 days to object | OpenAI DPA |
| Anthropic | 15 days to object, or the customer "is deemed to consent" | Anthropic DPA |
| xAI | "at least fifteen (15) days before such changes take effect," and 15 days to object | xAI DPA |
| Mistral | Notice "prior to engaging such Subprocessor"; 10 days to object | Mistral DPA |
Microsoft's own timeline shows how fast an AI subprocessor can arrive. It added OpenAI to its list on June 23, 2026, made it available on July 9, and turned OpenAI-operated models on for all users at eligible commercial customers on July 24, unless an administrator switched them off (Microsoft).
How Morse limits what a subprocessor chain can keep
A subprocessor list is a map of who may touch data a company keeps. Each name on it is another place your chats can be copied, reviewed, or kept for a time, and the list for a consumer app is often one you can't read. Morse is built to have little to pass along. Your chat history is stored in your browser, encrypted, and Morse's servers don't keep your messages or log what you write.
- Private, the default, for chats no provider should keep. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing of what you send. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't. Open-weight models such as DeepSeek, Kimi, and GLM, and xAI's Grok, run here.
- E2EE, for chats no company in the chain should read. Your browser checks the enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay ciphertext they can't read. The enclaves are run by third-party providers, not by Morse, and the guarantee depends on the enclave hardware and attestation working as designed. E2EE comes with Pro, at $19 a month.
- Anonymous, for GPT, Claude, and Gemini without your name attached. Morse's proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider still sees the words, and may keep them for a limited time to check for abuse under its own policy, so leave identifying details out.
Morse doesn't train AI on your content, sell your data, or show ads, and it runs no advertising trackers or third-party analytics. Because we don't store chats, we usually have very little to give authorities. Clearing your browser deletes your history for good, and we can't recover it, so keep a copy of anything you need. Our privacy policy says what the account itself holds.
Start a chat from the Morse home page without an account, or sign up free with no card.
Common questions
Is a subprocessor allowed to use my chats for its own purposes?
Not under the contracts these lists rest on. OpenAI's list, for example, says its moderation subprocessors "only process the content to assist OpenAI in its review," and Microsoft's Privacy Statement says its vendors "may not use your data for other purposes."
Can I get notified when a list changes?
Usually yes, even without a contract. OpenAI's list has a sign-up form for new subprocessor notices, and Anthropic's and Mistral's trust centers have a subscribe button. Only business customers with a data processing agreement get the right to object.
Does a subprocessor list tell me where my chats are stored?
Only roughly. The lists name countries where a vendor may process data, often several per vendor, and some say only "Worldwide." Where your own chats sit depends on the product and your region; our guide to where AI chat data is stored covers each app.
Why do AI companies list each other?
Because one company's app often runs another company's model. When Perplexity answers with Claude, or Microsoft 365 Copilot answers with a model from OpenAI or Anthropic, the model maker processes that request as a subprocessor.