Acuity AI Advisory
← Insights
·7 min read

Is Claude Safe for Client Data? GDPR, Retention and the Settings That Matter

G

Ger Perdisatt

Founder, Acuity AI Advisory

Claude can be used with client data on a business plan (Team or Enterprise), where your firm is the controller, Anthropic is the processor under its Data Processing Addendum and your content is not used for model training by default. Personal Free, Pro and Max accounts should not be used for client work.

Partners and practice managers ask us some version of the same question before any Claude rollout: can we put client material into it?

The short answer is yes, on the right plan and with a few settings checked. This post goes through the account type, the contract, retention and your own GDPR records. It is general information, not legal advice, and your data protection adviser should review the final position for your firm.

Personal accounts and business accounts are different products

Anthropic runs two sets of terms. The individual Free, Pro and Max plans are consumer products. Claude Team, Claude Enterprise and the API are commercial products. The difference matters more than any setting.

On a consumer plan, each person controls a model improvement setting in their own privacy settings. If they allow it, Anthropic says it may keep their chats in a de-identified form for up to five years in its training pipelines. The firm has no view of what is in those accounts and no way to switch the setting off for them.

On a commercial plan, Anthropic says it does not use your inputs or outputs to train its models by default. The exception is its Development Partner Program, which a customer has to choose to join.

So the first policy line is simple: personal Claude accounts are not for client work. If people have been using them, move them to a business plan and ask them to stop.

Who is controller and who is processor

Under Anthropic's commercial terms, your firm is the controller of the data your users submit, and Anthropic acts as processor on your behalf. Anthropic's own help article says it processes that data only on the customer's instructions, to provide the Claude service, and that the customer can access and export the data its users submit.

The Data Processing Addendum, with Standard Contractual Clauses, is built into the Commercial Terms of Service, so accepting the terms means accepting the DPA. Save a copy with your other processor agreements.

Where the data goes

This is the point Irish firms most often get wrong in both directions.

Anthropic's privacy centre says it uses several cloud providers, and that by default customer traffic may be routed to select countries in the US, Europe, Asia and Australia. It also says plainly that data is stored in the US. Data may be processed in other countries where Anthropic operates, for things like safety review, support and incident response.

There is no EU-only storage option on Claude Team today. Enterprise customers on usage-based billing have a US-only processing setting, which narrows routing rather than keeping it in Europe. If your firm has a hard rule that client data must stay in the EU, Claude Team will not meet it, and you should take that question to your adviser before you buy.

For most professional firms the transfer is handled the same way as other US cloud services they already use: a processor agreement with Standard Contractual Clauses, recorded in the firm's records and covered by a transfer assessment.

Retention: what Anthropic keeps and for how long

For commercial customers, Anthropic's privacy centre says:

  • Chats are kept in the product so people can return to them.
  • When someone deletes a chat, it leaves their history straight away and is deleted from Anthropic's back-end storage within 30 days.
  • Content flagged under its usage policy can be kept for up to two years, and the safety classification scores for up to seven years.
  • Feedback submitted through the thumbs up or down buttons is stored for up to five years with user and customer IDs removed, and may be used for training.

The last point catches firms out. Feedback is the one route by which business content can reach training. On Team and Enterprise, an Owner can turn feedback off for the whole organisation under Organization settings, Data and Privacy, using the "Rate chats" setting. For a firm handling client files, switch it off.

On Enterprise only, Owners can set a custom retention period for chats, projects and artifacts. The minimum is 30 days. Without a custom period, Anthropic says data is kept indefinitely. Team does not have custom retention controls.

What Team and Enterprise give an administrator

Claude Team suits most Irish firms of 10 to 100 people. On its pricing page Anthropic lists, for Team, single sign-on, central billing and administration, domain verification, usage analytics, admin controls for connectors and no model training on your content by default.

Enterprise adds the controls a compliance or IT function will ask about:

  • audit logs
  • the Compliance API, for monitoring and observability
  • custom data retention
  • role-based access with fine-grained permissions
  • SCIM user provisioning
  • IP allowlisting and network-level access control
  • spend limits per user and for the organisation

Enterprise has a 20-seat minimum and charges usage at API rates on top of the seat price. Our Claude pricing guide sets out what each plan costs a typical Irish team. AI at Work, part of Acuity, also has a plain comparison in Claude Team or Claude Enterprise.

Connecting Microsoft 365

Much of Claude's value at work comes from the Microsoft 365 connector, which lets it search Outlook, SharePoint, OneDrive and Teams. Anthropic's documentation says the connector uses delegated permissions, so it acts as each signed-in person and reaches only what they could already open. A Global Administrator in your Microsoft tenant has to consent before anyone can connect, and on Team and Enterprise a Claude organisation Owner has to enable it too. Write tools, which let Claude send email or edit files, are only available if an administrator turns them on.

The risk here is not Claude. It is SharePoint sites shared more widely than anyone meant. Review sharing before you connect. The AI at Work guide is it safe to connect Claude to Microsoft 365? covers the checks.

What to record in your GDPR records

Add Claude to your records of processing in the same way as any other processor. For most firms that means:

  • Anthropic as a processor, with the plan you use and the date you accepted the commercial terms and DPA
  • the categories of personal data staff may enter, and any categories they may not
  • the transfer to the US, the Standard Contractual Clauses relied on and your transfer assessment
  • the retention position, including the 30-day deletion window and any Enterprise retention setting
  • which connectors are enabled and who approved them
  • a note that feedback submission is switched off, if you have done so

If staff will enter special category data, such as health information in a personal injury file, check with your adviser whether a data protection impact assessment is needed before they start.

Sensible policy lines

Keep the Claude section of your AI policy to a page. The lines that matter most are:

  • Use only the firm's Claude Team or Enterprise account for work. Personal accounts are not to be used for client material.
  • Do not enter special category data or highly sensitive client material unless a partner has approved that use.
  • New connectors are approved by a named person before they are switched on.
  • Feedback buttons are off at organisation level.
  • Anything that goes to a client is checked by a person against the source documents first.

If you already have an AI policy, add these rather than starting again. Our AI governance advice covers the wider policy and records work, and the post on rolling out Claude company-wide sets out the order to do things in.

If you would like help setting this up for your firm, along with training on real client work, tell us about your team.

claudeai governanceirelandprofessional servicesai policy