This is a plain-English summary of published guidance, not legal advice.
Is ChatGPT GDPR compliant? The short answer
Free ChatGPT isn't fit for client data. ChatGPT Business, Enterprise and the API can be used in a UK GDPR compliant way, if your firm signs the DPA, does a DPIA and sets a policy.
No AI tool is "GDPR compliant" on its own. Because your firm is the data controller for clients' personal data, compliance depends on how you configure and govern the tool. The vendor determines whether compliant use is possible through three terms: whether they train on your prompts, whether they will sign a data processing agreement (DPA) with your firm, and where data is stored and processed.
The same split applies across all four vendors below: personal plans (ChatGPT Free and Plus, Claude Pro, Copilot on a personal Microsoft account) are consumer products, whereas business plans are designed for organisations. Staff pasting client emails into personal accounts is shadow AI, and it carries far more risk than a properly configured business tier.
ChatGPT, Copilot, Claude and Otter compared
Each row covers one product tier, and each cell links directly to the vendor's documentation (checked November 2026).
What the columns mean
- Trains on your data. If it does, what staff type can end up shaping a model other people use. That's hard to square with client confidentiality.
- DPA. UK GDPR Article 28 needs a written contract with any processor handling personal data for you. The ICO explains what goes in it in its guidance on controller and processor contracts. A personal account has no contract with your firm at all.
- Data residency. Data leaving the UK isn't banned. A transfer needs a lawful route, like the UK-US data bridge for certified US companies or the safeguards in the ICO's international transfers guidance. UK storage just makes the DPIA shorter.
- SOC 2 and ISO 27001. These show the vendor has audited security controls, though they do not on their own make your use of the tool GDPR-compliant.
Is ChatGPT Business GDPR compliant?
It can be used compliantly. OpenAI says it doesn't train on Business, Enterprise, Edu or API data by default, and it will sign a DPA (OpenAI business data). Business has a SOC 2 Type 2 report.
What it doesn't have is UK storage. OpenAI's data residency covers Enterprise, Edu and the API, where eligible new workspaces can keep data at rest in the UK or Europe. On Business, your DPIA has to cover the transfer. OpenAI lists its ISO 27001, 27017, 27018 and 27701 scope as the API, Enterprise and Edu on its trust portal.
On Free, Plus and Pro, the "Improve the model for everyone" setting is on by default (OpenAI data controls FAQ). Turning it off stops training, but there's still no DPA with your firm, so it doesn't fix the problem.
Is Copilot GDPR compliant?
Microsoft 365 Copilot and Copilot Chat on a work account are the simplest option for a firm already on Microsoft 365. Microsoft says prompts, responses and data from Microsoft Graph "aren't used to train foundation LLMs", and Copilot is covered by the same Product Terms and DPA as the rest of Microsoft 365 (Microsoft Learn).
Four things to check before you switch it on:
- Permissions. Copilot can surface anything a user has access to. If your SharePoint folders are open across the firm, Copilot will find the partners' pay review. Tidy permissions first.
- Web search (Bing grounding). When web search is enabled, Copilot turns your prompt (and, when you prompt inside an open Office file or reference a document, terms drawn from that document) into short search queries sent to Bing. Microsoft states that neither the DPA nor the EU Data Boundary applies to those generated search queries, because Bing runs separately from Microsoft 365 with Microsoft acting as an independent data controller (Microsoft Learn). Many regulated firms switch web search off, or disable it in Work mode, using the "Allow web search in Copilot" admin policy.
- Where it's processed. EU traffic stays in the EU Data Boundary. Microsoft says customers outside the EU "may have their queries processed in the US, EU, or other regions".
- Anthropic models. If your admin turns on Claude models inside Copilot, Microsoft notes they are processed outside the EU Data Boundary.
Copilot on a personal Microsoft account is different. It can use conversations for training unless you switch off "Training on conversation activity" (Microsoft support), and there's no contract with your firm.
Is Claude AI GDPR compliant?
Claude Team, Enterprise and the API run under Anthropic's Commercial Terms, which include a DPA and say "Anthropic may not train models on Customer Content from Services". Anthropic lists SOC 2 Type II, ISO 27001 and ISO 42001 for these products (Anthropic certifications).
The main limitation is data residency. On Anthropic's own platform, workspaces are US-only and processing can be set to US or global rather than UK or EU (Claude data residency docs). Your DPIA needs to cover that.
Claude Free, Pro and Max are consumer plans. Since Anthropic's 2025 consumer terms update, chats are used for training if the setting is on, and kept for up to five years. With it off, the 30-day retention applies. Either way, there's no DPA with your firm.
Is Otter AI GDPR compliant?
Meeting bots are where most firms get caught out, because the bot records people who never agreed to it. Otter's privacy and security page says it de-identifies user data "before training our models", and it doesn't describe a way to opt out. Its privacy policy says data is hosted with US cloud providers, keeps it "as long as necessary", and relies on the UK extension to the Data Privacy Framework for transfers.
It has a SOC 2 Type 2 report. Its policies are "based on" ISO 27001, which isn't the same as being certified. For client calls, review the DPA properly and tell everyone on the call before the bot joins.
Is it safe for staff to use ChatGPT with client data?
Only if the firm has set it up. Before anyone puts client data into an AI tool:
- Approve a business tier from the table and restrict personal or unapproved tools.
- Sign the vendor's DPA and keep a copy with your records of processing.
- Do a DPIA. The ICO's AI guidance says most uses of AI with personal data will need one.
- Update your privacy notice and engagement letters to say you use AI and who the processor is.
- Write a short AI policy covering approved tools, permitted client data and output review.
- Require a staff member to review outputs before they reach a client.
To see which of these your firm already has in place, score it against our AI readiness checklist.
Professional regulatory duties apply alongside UK GDPR. The SRA's August 2026 warning notice on misuse of AI says client information should only go into AI systems with proper safeguards in place, and warns of a "permanent waiver of privilege". ICAEW's November 2026 note on AI and client confidentiality says the same for accountants. More in our SRA AI guidance and ICAEW AI guidance summaries.
What the ICO says
The ICO hasn't approved or banned any AI product. Its guidance applies to all of them:
- Guidance on AI and data protection: fairness, transparency, accuracy and security when you use AI with personal data.
- Its response to the generative AI consultation: the ICO's positions on lawful basis, purpose limitation and accuracy for generative AI.
- DPIAs, processor contracts and international transfers: the three requirements behind the comparison table above.
The alternative: private AI on your own systems
A business tier solves the contract, but out of the box it still won't know your precedents, clients or house style. We build private AI workflows that connect to your existing systems, run on your data and keep human sign-off on client deliverables. You own the software and data from day one.
- Private AI for your firmAI workflows running inside your own systems, on your data, with human sign-off.
- AI policy templateA free AI policy template covering approved tools, client data rules and output review.
- AI readiness assessmentEnter your website URL to see which workflows across your services AI can already handle.
