Is it GDPR-safe to pass client data between AI assistants?

The short answer

It can be, but it depends on roles and contracts, not on the tool alone. Before client data moves between AI assistants, you need to know who is the controller, who processes the data for you, where it is stored, how long it stays, and whether each provider has signed a data processing agreement with you.

This guide explains the roles in plain words, shows the three parties in every AI handoff and gives you an eight-point checklist to run before your team starts. At the end, we answer the same checklist for Colle.

Controller vs processor in plain words

GDPR gives the companies that touch personal data different roles. Two of them matter here.

The controller decides why personal data is processed and how. If your firm keeps a client's contact details and uses them to run a project, your firm is the controller for that data.

The processor handles personal data on the controller's behalf, and only on its instructions. A hosting company, an email provider or a tool that stores your files usually acts as a processor for that content.

Sometimes your firm is itself the processor. An agency that works with its client's customer list processes that list for the client. The client is the controller, and new tools in the chain may need the client's approval. Read your client contracts before you add an AI tool to the work.

GDPR requires a written contract between a controller and a processor (Article 28). In practice, this is the data processing agreement, or DPA. It sets out what the processor may do with the data, the security it must keep, how it helps you answer people's requests and what happens to the data when the contract ends.

A processor can use other processors. These are sub-processors. The controller has to authorise them, usually through a list attached to the DPA, and the processor stays responsible for them.

One company can hold both roles at once. A software vendor is often the controller for its own website visitors, accounts and invoices, and a processor for the content its customers put into the product. Ask each vendor which role it takes for which data.

Three parties in every AI handoff

When someone says to their assistant “send this contract to Rachel”, the data passes through more hands than it seems. There are three parties.

1. Your company

Your company decides to use AI assistants for client work and chooses the tools. For your clients' personal data, you are the controller, or a processor acting for your client. Either way, you answer for the choice of tools.

2. The assistant provider

Claude is run by Anthropic and ChatGPT by OpenAI. When someone pastes a client document into a chat, the provider processes it under the agreement your company has with them. Check which plan your team is on, read its terms and data processing terms, and make sure staff do not use private accounts for client work.

3. The handoff tool

If the work moves to a colleague through a separate tool, that tool stores the summary, the facts and the files between the two assistants. It is another company, with its own servers, its own retention rules and its own staff. It needs its own DPA with you.

A common gap is to check only the assistant provider. A handoff tool exists to keep the case for the next person, so it holds client data on purpose. Check it with the same care.

If your handoffs happen today by email, a chat app or a shared drive, those are parties too. Moving to AI handoffs does not create the question. It makes it visible.

The checklist: eight questions before client data moves

Ask these questions of every tool in the chain: the assistant provider and the handoff tool. Write the answers down. They also help with your own records of processing.

Some of these points need a closer look.

Where the data is stored

EU hosting makes things simpler, but it is not the whole answer. Check who runs the hosting and whether data leaves the European Economic Area for support or backups. Transfers outside the EEA need a legal basis under Chapter V of GDPR, and your lawyer will want to see it.

Retention and erasure

“We delete data on request” is not enough. Ask for numbers: how many days after the contract ends, and how many more until the backups are gone. Ask how one client contact's data is removed when that person asks, without deleting the whole case.

Training use

Look for a clear sentence in the contract or the privacy policy, not a setting on a help page. If the answer differs by plan, check the plan your team pays for.

Access controls

Many handoff risks are ordinary ones: a contract sent to the wrong Tom, or a whole team seeing work meant for two people. Check whether you can limit a piece of work to named people, and whether the tool asks before it guesses a recipient.

How Colle answers each

Here is the same checklist for Colle. Every answer below comes from our published security and data facts.

Where is the data stored?

In the EU. Colle is hosted in Frankfurt on Fly.io, with encrypted disks and daily backups. Fly.io keeps the data in Frankfurt; the company itself is based in the US. Data is encrypted in transit (TLS) and at rest.

Is there a DPA?

Yes. For customer content (summaries, facts, files and cases), Colle is your processor under a DPA. For our own website, accounts and billing, Colle is the controller. To get the DPA, email michael@trycolle.com with the subject “Colle DPA request”. EU-only data and the DPA are part of the one plan; see pricing.

Who are the sub-processors?

Claude (Anthropic) and ChatGPT (OpenAI) are not Colle's sub-processors. Your conversation is processed by the assistant provider under your own agreement with them. Colle's hosting provider is Fly.io, as above.

How long is the data kept?

Customer data is deleted within 30 days after your subscription ends, and from backups within the following 14 days. Security logs are kept for 30 days. An administrator can export or delete your company's data at any time.

Can you erase a person's data?

Yes. Personal data in Colle is encrypted with a separate key for each person. Erasing a person destroys that key.

Is the data used for training?

No. Colle does not use personal data or customer data to train AI models, and it does not sell personal data.

Who can see what?

Each company's data is isolated with row-level security on every query. A handoff sent to a person is seen only by the sender and the recipients. Work sent to a space notifies the space's members. Cases in a closed space are seen only by its members; cases in an open space are visible to the whole company. Personal data is not passed to anyone without access. Text from a colleague is marked as untrusted and scanned for prompt injection.

Is there an audit log?

Yes. The audit log is tamper-evident: its entries are linked in a hash chain. Administrators can export it.

If something goes wrong

In the event of a personal data breach, Colle notifies your organisation without undue delay, because for your content it is the processor; you, as the controller, decide whether to notify the supervisory authority within 72 hours (GDPR Articles 33–34). For its own website, accounts and billing, where Colle is the controller, it notifies the authority itself.

More detail is on the security page. To see how a handoff travels from one person's assistant to another's, read how Colle works.

What stays with you

Colle answers for Colle. Your agreement with Anthropic or OpenAI, the plans your staff use and your own records of processing stay with your company. Sending handoffs from ChatGPT needs ChatGPT Business or Enterprise; see the ChatGPT integration.

FAQ

Is Claude or ChatGPT GDPR-safe on its own?

A tool alone does not settle it. What matters is how your company uses it: the plan, your agreement and DPA with the provider, the settings, and what staff put into it. Check the terms of the plan your team actually uses.

Do we need a DPA with the handoff tool if we have one with the assistant provider?

Yes. They are separate companies with separate servers. Each company that processes personal data for you needs its own agreement. Colle provides its DPA on request.

Does EU hosting alone make a tool GDPR-safe?

No. Location is one of the eight questions. You still need a DPA, clear retention, erasure, access controls and an audit log.

Can we keep client data out of the assistants altogether?

You can reduce it. Agree simple team rules: which client data may go into a chat, which plan staff must use, and who may receive what. GDPR asks you to use only the data you need, and fewer copies in fewer places is the easiest way to start.