“Where do my customers’ emails go?” It’s the right question. The answer depends on a document few people read: your contract with the supplier. On 2 November 2026 we read the official pages of Anthropic, OpenAI, Google and Microsoft, and we say which ones are missing. Terms change, so read them again before you decide.
Personal account or business contract
If someone in your office pastes an email into a personal account, consumer rules apply. And personal doesn’t mean free.
Anthropic counts its paid Pro and Max plans as consumer plans. Its European terms say it may train its models on your content unless you opt out in your settings. Even then, that leaves anything you send as feedback and any conversation flagged for a safety review.
For the Gemini apps, Google says it uses your activity to train its models, among other things, and that human reviewers read some chats. You can switch the setting off: from then on chats no longer go to improving the models, apart from feedback you send, but Google still keeps them for 72 hours and human reviewers may read them for safety reasons. Chats a reviewer has already read stay for up to three years. Google asks you not to enter confidential information.
Microsoft has two pages for personal Copilot. The one for the older app says it uses conversations for training, apart from some exceptions and anyone who opts out in settings, and normally keeps them for 18 months. For the web and the new app, available since 18 August 2026, it points to another page: check there.
For personal ChatGPT plans, check your account settings: OpenAI’s official page wouldn’t open for us on 2 November 2026, and we only write what we have read.
A business contract or the API, the direct link between a system and the model, changes the picture.
Anthropic’s commercial terms say it may not train models on customer content. OpenAI says the same of its API, unless you opt in, and by default of ChatGPT Business and Enterprise. Google says it of the paid Gemini API, in Europe also of AI Studio and the API’s free quota, and, unless you give permission, of Workspace. Microsoft says it of Microsoft 365 Copilot and of the models it sells on Azure.
At Anthropic, even on a business plan, a thumbs up or down can bring that conversation into training. On Team and Enterprise plans, the account owner can turn those buttons off for everyone.
How long it stays, and where
Not training isn’t the same as not keeping. On the API, Anthropic deletes within 30 days unless agreed otherwise or required by law. If its automated systems flag a breach of its usage rules, it can keep data for up to two years, and thumbs-up or thumbs-down feedback for up to five. On the API, OpenAI normally keeps abuse-monitoring logs for up to 30 days, unless the law requires longer or it’s needed to prevent harm. For the API, Google’s terms speak of a limited period; its API usage policies say 55 days, for abuse monitoring and any legal disclosures.
For its business products, Anthropic may by default route traffic to the US, Europe, Asia and Australia, unless agreed otherwise or on your instructions, and stores data in the US. OpenAI has a European region, but you have to ask, and conditions apply. On its business cloud platform (Gemini Enterprise Agent Platform), Google offers an access point that keeps processing inside the EU, and says its global one gives no guarantee of location. Microsoft keeps European Microsoft 365 Copilot traffic within its EU Data Boundary. Anthropic’s models stay outside it, and for European customers they’re off until an admin turns them on.
What the GDPR asks
If those emails hold people’s details, you decide why and how they’re used: you’re the controller. A supplier handling the data on your behalf is the processor: it acts only on your instructions, under a contract. On a personal account that may not be so, because the supplier may decide for itself. Ask your lawyer how your case fits. If the processor brings in another supplier, it needs your written authorisation, specific or general, and you must be able to object.
If the data leaves the European Economic Area, you need an adequacy decision or safeguards such as standard contractual clauses. For the US, as of November 2026, the Commission’s decision covers only organisations taking part in the Data Privacy Framework. People must be told who receives their data.
Models on your own premises
A model can also run on your own machine. Ollama, one tool for doing so, says it doesn’t see your prompts or data when you run locally. It also offers cloud models, and there it does process your prompts.
OpenAI’s guide to its open models gives about 16 GB of video memory for the smaller one and, as the best choice, at least 60 for the larger. The guide is archived, so treat that as a rough scale.
On your own premises the supplier is you: updates, security and backups. Whether the model is good enough, only a test on your real jobs will tell.
Five questions for anyone
Put them to any supplier, us included. Ask for the answers in writing.
- Do you use my data to train models? Where does it say so?
- How long do you keep it, and with what exceptions?
- Which country is it stored in, and which is it processed in?
- Who can read it, both people and other suppliers?
- Will you give me a data processor contract, and what’s left when I stop?
For the legal side, check with your lawyer. Our commitment is already on the site: “The emails and the data stay yours. To process them they pass through outside providers, such as whoever runs the model: before anything is connected, we tell you in writing where they go and who can see them, and you decide.”
Sources
- Anthropic, Commercial Terms of Service, in force from 17 June 2025, read on 2 November 2026. www.anthropic.com
- Anthropic, Consumer Terms of Service (consumers in the EEA and Switzerland), in force from 8 October 2025, read on 2 November 2026. www.anthropic.com
- Anthropic, Is my data used for model training? (commercial products), Privacy Center, updated 18 August 2026, read on 2 November 2026. privacy.claude.com
- Anthropic, Is my data used for model training? (consumer plans), Privacy Center, updated 16 March 2026, read on 2 November 2026. privacy.claude.com
- Anthropic, How long do you store my organization’s data?, Privacy Center, updated 1 July 2026, read on 2 November 2026. privacy.claude.com
- Anthropic, Where are your servers located?, Privacy Center, updated 15 June 2026, read on 2 November 2026. privacy.claude.com
- Anthropic, Privacy Policy, in force from 10 September 2026, read on 2 November 2026. www.anthropic.com
- OpenAI, Data controls in the OpenAI platform, developer documentation, read on 2 November 2026. developers.openai.com
- OpenAI Academy, Data governance and compliance, updated 17 September 2026, read on 2 November 2026. academy.openai.com
- OpenAI, How to run gpt-oss with vLLM, Cookbook (archived recipe, undated), read on 2 November 2026. developers.openai.com
- Google, Gemini API Additional Terms of Service, in force from 23 March 2026, read on 2 November 2026. ai.google.dev
- Google, Abuse monitoring, Gemini API, updated 9 June 2026, read on 2 November 2026. ai.google.dev
- Google, Gemini Apps Privacy Hub, updated 26 October 2026, read on 2 November 2026. support.google.com
- Google, Generative AI in Google Workspace Privacy Hub, updated 6 October 2026, read on 2 November 2026. knowledge.workspace.google.com
- Google Cloud, Data residency, Gemini Enterprise Agent Platform, undated, read on 2 November 2026. docs.cloud.google.com
- Microsoft, Data, Privacy, and Security for Microsoft Copilot, Microsoft Learn, updated 18 August 2026, read on 2 November 2026. learn.microsoft.com
- Microsoft, Data, privacy, and security for Foundry Models sold by Azure, Microsoft Learn, updated 19 May 2026, read on 2 November 2026. learn.microsoft.com
- Microsoft, Privacy FAQ for Microsoft Copilot, Microsoft Support, read on 2 November 2026. support.microsoft.com
- Microsoft, Microsoft Copilot for individuals: your data, privacy, and responsible AI, Microsoft Support, read on 2 November 2026. support.microsoft.com
- Microsoft, Anthropic models in Microsoft Online Services, Microsoft Learn, updated 18 September 2026, read on 2 November 2026. learn.microsoft.com
- European Data Protection Board (EDPB), SME data protection guide: data controller and data processor, read on 2 November 2026. www.edpb.europa.eu
- EDPB, SME data protection guide: international data transfers, read on 2 November 2026. www.edpb.europa.eu
- EDPB, SME data protection guide: respect individuals’ rights, read on 2 November 2026. www.edpb.europa.eu
- European Commission, Adequacy decisions, read on 2 November 2026. commission.europa.eu
- Ollama, FAQ, official documentation, undated, read on 2 November 2026. docs.ollama.com